# Slashspace > Slashspace (formerly Rabbitholes AI) is a canvas-first AI interface for hands-on, long-context work on hard problems. It's built for researchers, learners, strategists and writers who want to think with AI, not hand their thinking over to it. Slashspace is a desktop app for macOS, Windows and Linux. Most AI apps aim for an away-from-keyboard experience: ask, wait, accept. That suits grunt work. Discovery work (research, learning, strategy, and producing high-quality work that doesn't read like AI) needs the person hands-on, thinking in branches, iterations and variations. Slashspace puts that work on an infinite canvas. Every chat is a node: fork a conversation to chase a tangent without polluting the main thread, compare answers from different models side by side, and connect exactly the sources and notes each chat should see. ## What Slashspace does - Branch conversations on a canvas: fork any chat to explore a tangent while the main thread stays clean - Compare iterations and variations: run the same question through several models or prompts side by side - Wire in sources: PDFs, documents, web pages, YouTube videos, images and notes become context you connect to chats - Hold long context across a project: a canvas keeps weeks of work in view, with every chat keeping its connected context - Use the models you already pay for: hosted models, your own API keys, your Claude or ChatGPT subscription through Claude Code and Codex, or local models with Ollama - Canvas Orchestrator: an agent that creates, connects and updates nodes and delegates work to subagents, with the person approving each switch into Orchestrator mode ## What makes it different - Built for human-mind-centric, hands-on work rather than away-from-keyboard automation - Conversation branching on the canvas is the core of the product - Local models (Ollama) and existing Claude or ChatGPT subscriptions work alongside hosted models and bring-your-own API keys - Flat pricing: chats on your own keys, subscriptions and local models cost nothing extra - Local-first: canvases, conversations and files are saved in a folder on your own computer - Slashspace does not compete with linear chat apps like ChatGPT, which suit quick questions and grunt work; you can use their models inside Slashspace - Competitors: Heptabase, Anara, Arky, Poppy AI ## Who uses it - Researchers and academics working through literature, experiments and long investigations - Students and self-learners taking on a hard subject one branch at a time - Strategists, founders and business owners researching markets, competitors and decisions - Writers and creators producing high-quality work that shouldn't read like AI - Polymaths connecting ideas from distant fields - Used by people at Qualcomm, HubSpot, Google, OpenAI, UC Santa Barbara, Texas A&M University and UC Riverside - 2,500+ users ## Company - Name: Slashspace (formerly Rabbitholes AI), operated by Lokus LLC - Founded: 2024 as Rabbitholes AI; renamed Slashspace as the product vision expanded - Founder: Praneeth Pike - Headquarters: New York, NY, USA; the company operates remotely - Support: support@slashspace.ai and Discord, with replies within a day - [About Slashspace](https://www.slashspace.ai/about): company, team, key facts and FAQ - [Contact](https://www.slashspace.ai/contact): support email, Discord, reply times and mailing address ## Pricing - Slashspace is a subscription: $89 a year, with every update while you subscribe. It renews every year. - Cancel any time from the Stripe billing portal on your account page. Cancelling keeps access to the end of the paid year. - Slashspace Lifetime: $249 one-time, with lifetime updates. No subscription. - There is no free trial. Both plans are refundable within 7 days, no questions asked. - To buy, enter your email on the website, answer three quick questions and pay on Stripe. You don't need an account first. Then download the desktop app and sign in with that same email. - Customers who bought the earlier $89 one-time license keep it. When their updates window ends they can subscribe for $89 a year to keep getting updates. - The optional hosted wallet (Slashspace Wallet credits, 1 credit = 1 cent) pays for Slashspace's hosted AI models, including Auto, and AI tools such as image generation. Both plans include a one-time grant of 500 credits. A zero wallet balance does not block app access, BYOK, subscription providers, or local models. - Each plan covers 2 devices. - 7-day no-questions-asked refund policy - [Full pricing details](https://www.slashspace.ai/pricing) ## Docs - [Quickstart](https://www.slashspace.ai/docs/index.md): Install, sign in, ask a first question, branch it, and bring in a source - [Why a canvas](https://www.slashspace.ai/docs/why-a-canvas.md): Why an AI canvas beats a chat thread for long-context exploratory work - [Core concepts](https://www.slashspace.ai/docs/core-concepts.md): Canvases, nodes, connections as context, branches and chat modes - [Canvases and folders](https://www.slashspace.ai/docs/canvas/canvases-and-folders.md): The sidebar, folders, and where canvases live on disk - [Navigating the canvas](https://www.slashspace.ai/docs/canvas.md): Pan, zoom, mini map, node navigator and canvas settings - [Adding nodes](https://www.slashspace.ai/docs/canvas/creating-nodes.md): Toolbar, hotkeys, right-click menu, paste and drag-and-drop - [Connections and context](https://www.slashspace.ai/docs/canvas/connections.md): Control exactly what each chat reads - [Branching](https://www.slashspace.ai/docs/canvas/branching.md): Fork any answer into a new branch - [Organizing the canvas](https://www.slashspace.ai/docs/canvas/organizing.md): Selection, alignment, groups, Tidy Up and undo - [Search](https://www.slashspace.ai/docs/canvas/search.md): Local Search and AI Search - [Side view](https://www.slashspace.ai/docs/canvas/side-view.md): Read and work in one node at full height - [Export](https://www.slashspace.ai/docs/canvas/export.md): Download a canvas as Markdown or JSON ## Nodes - [Chat node](https://www.slashspace.ai/docs/nodes/chat-node.md): AI conversations with modes, models, tools and message actions - [Text node](https://www.slashspace.ai/docs/nodes/text-node.md): Rich-text notes as context - [Web node](https://www.slashspace.ai/docs/nodes/web-node.md): Any public web page as context - [Post node](https://www.slashspace.ai/docs/nodes/post-node.md): YouTube, TikTok, Instagram, Facebook, X and LinkedIn transcripts as context - [Document node](https://www.slashspace.ai/docs/nodes/document-node.md): PDF, Office, text and code files as context - [Image node](https://www.slashspace.ai/docs/nodes/image-node.md): Images as context - [Scratch pad](https://www.slashspace.ai/docs/nodes/scratch-pad-node.md): Sketches that connected chats see as images - [Group](https://www.slashspace.ai/docs/nodes/group-node.md): Group related nodes and use them as context ## Work with AI - [Choosing a model](https://www.slashspace.ai/docs/ai-features/models.md): Hosted models, subscriptions, API keys, local models, reasoning effort and defaults - [Agent mode and tools](https://www.slashspace.ai/docs/ai-features/agent-mode.md): Connectors, tool calls and approvals - [Sub-agents](https://www.slashspace.ai/docs/ai-features/sub-agents.md): Delegating work to child chat nodes - [Orchestrator mode](https://www.slashspace.ai/docs/ai-features/orchestrator.md): Give a goal and let the AI build the canvas - [Image generation](https://www.slashspace.ai/docs/ai-features/image-generation.md): Image mode, reference images and cost - [Slash commands, skills and prompts](https://www.slashspace.ai/docs/ai-features/slash-commands.md): Reusable prompts and skills from the / menu - [Personas](https://www.slashspace.ai/docs/ai-features/personas.md): Saved system prompts for any chat - [Voice input](https://www.slashspace.ai/docs/ai-features/voice-input.md): Dictation in chat ## Connect models and tools - [Subscription providers](https://www.slashspace.ai/docs/settings/subscription-providers.md): Claude Code, Codex and Cursor on your existing subscription - [API providers](https://www.slashspace.ai/docs/settings/api-providers.md): OpenAI, Anthropic, Google and more with your own API keys - [Ollama](https://www.slashspace.ai/docs/settings/ollama.md): Local models - [Custom providers](https://www.slashspace.ai/docs/settings/custom-providers.md): Any OpenAI-compatible endpoint - [Connectors](https://www.slashspace.ai/docs/settings/connectors.md): Connect the apps your chats can use - [Custom MCP connectors](https://www.slashspace.ai/docs/settings/mcp-servers.md): Your own tools via MCP - [Agent skills](https://www.slashspace.ai/docs/settings/skills.md): Install skills from the skills.sh registry - [Add to Slashspace button](https://www.slashspace.ai/docs/settings/add-to-slashspace.md): One-click install links for MCP server authors ## Account and help - [Plans and billing](https://www.slashspace.ai/docs/plans-and-billing.md): Yearly subscription, Lifetime, refunds and the hosted wallet - [Account and devices](https://www.slashspace.ai/docs/settings/account-management.md): Account settings and device slots - [Privacy and your data](https://www.slashspace.ai/docs/privacy.md): What stays on your computer and what leaves it - [Keyboard shortcuts](https://www.slashspace.ai/docs/canvas/keyboard-shortcuts.md): Every shortcut by platform - [Troubleshooting and support](https://www.slashspace.ai/docs/troubleshooting.md): Common problems and how to get help ## API - [OpenAPI spec](https://www.slashspace.ai/openapi.json): The website's public endpoints (docs search, docs as markdown, site indexes). The homepage, About, Contact, Pricing and Download pages also return markdown for `Accept: text/markdown` ## Full Documentation - [Complete docs in plain text](https://www.slashspace.ai/llms-full.txt): Full product documentation, blog posts and FAQs compiled for LLM ingestion ## Legal - [Terms of Use](https://www.slashspace.ai/legal/terms): Usage terms and refund policy (7 days, no questions asked) - [Privacy Policy](https://www.slashspace.ai/legal/privacy-policy): How data is handled ## Optional - [Changelog](https://www.slashspace.ai/changelog): Release notes and update history - [Migrating from Rabbitholes AI](https://www.slashspace.ai/docs/migration.md): What carries over from Rabbitholes AI - [Discord](https://links.rabbitholes.ai/discord): Community support and feature requests - [Email Support](mailto:support@slashspace.ai): Direct support contact --- # Full Documentation # Coming Soon... (https://www.slashspace.ai/docs/_web-app) No description available # Agent mode and tools (https://www.slashspace.ai/docs/ai-features/agent-mode) What the AI can do in Agent mode, how to choose the connectors a chat can use, and how tool calls and approvals work. In Agent mode, a chat node can act, not only answer. The AI can look things up in your connected apps, open other nodes on the canvas, and hand parts of a task to sub-agents, working through several steps before it replies. Every step shows up in the conversation, and changes to how the chat runs wait for you. Agent is the default for new chat nodes. Switch to **Chat** (`⌘2`, or `Ctrl+2` on Windows and Linux) when you want a focused reply with no tools, and back with `⌘1` (`Ctrl+1`). The [Chat node](/docs/nodes/chat-node#modes) page covers all the modes. {/* VIDEO: An Agent chat looks something up with a connector, then proposes switching to Image mode, and the image appears after Approve. */} ## What the agent can do - **Use your connectors.** Search your inbox, read a shared document, check a project board: whatever you've added in [Connectors](/docs/settings/connectors). OpenAI models on your own OpenAI key can also search the web. - **Read other nodes.** A chat with no incoming connections (and Summarized context on) gets a short summary of the canvas and can open any node to read it in full. Once you connect nodes, it reads only what's connected. See [Connections and context](/docs/canvas/connections#context-modes). - **Delegate to sub-agents.** It can split a task into parts, run each in a new connected chat node, and use their answers. See [Sub-agents](/docs/ai-features/sub-agents). - **Propose a mode or model switch** when your request needs one. Agent mode doesn't build your canvas. Apart from the chat nodes its sub-agents run in, it never adds, connects, moves or edits nodes. That's [Orchestrator mode](/docs/ai-features/orchestrator), and the agent offers to switch when you ask for nodes. ## Choose which connectors a chat can use Every connector that's connected and switched on in Settings is on in every Agent chat. To turn one off for a single chat: - Click **+** (tooltip "Tools"), open **Connectors** and uncheck it, or - Type `/`, open **Connectors** and pick it. Each entry shows "On in this chat" or "Off in this chat". The choice is saved with the chat node. **Manage connectors** opens Settings. In other modes the entry reads "Available in Agent mode". If the AI reaches for an app you haven't connected, a **Connect** card appears with an **Open Connectors** button. Connect the app, then send your message again. > **Note:** Connector tools run as soon as the AI calls them, with every model. Nothing asks first, so turn off connectors that can send or change things in chats that don't need them. ## Tool calls in the conversation Each tool call appears in the answer as a row with the tool's name and a one-line summary. Click it to see what came back; long results end in **Show more**, and failed calls show the error. Two or more calls in a row fold into one line such as "Used 3 tools". Click it to see each call. ## Approving tool calls Two kinds of call wait for you: proposals to switch the chat's mode or model, and tools that Claude Code or Codex ask permission for. The row asks "Allow this tool to run?" and the AI pauses: - **Approve** runs it, and the AI carries on. - **Reject** skips it. The row notes that the tool didn't run, and the AI continues without it. - **Always allow** appears in two places. On a Claude Code or Codex request, it approves that tool for the rest of the current reply; your next message asks again. On a proposed switch to Orchestrator mode, it also approves future Orchestrator switches. ## Mode and model switches When a request fits another mode or model better, the agent proposes a switch. Ask for an image and it proposes **Image** mode; ask it to create or connect nodes and it proposes **Orchestrator**; ask for a different model and it proposes that model. The proposal is a **Switch Mode** or **Switch Model** row with the target and a one-line reason. Nothing changes until you click **Approve**. The chat then switches and carries on with your request, so you don't need to resend it. **Reject** keeps the current setup. - Image mode needs a plan that includes image generation. Otherwise the switch is refused and an upgrade window opens. See [Image generation](/docs/ai-features/image-generation). - The agent can move a chat into Chat or Image mode but not back out, since those modes have no tools. Switching back is up to you. ### Auto-approve Orchestrator switches **Always allow** on an Orchestrator proposal turns on **Auto-switch into Orchestrator mode** in [Canvas settings](/docs/canvas#canvas-settings). While it's on, Orchestrator proposals apply without asking, for every canvas. Orchestrator mode sends canvas context to Slashspace's cloud, so turn the setting off if you'd rather approve each switch. ## Claude Code, Codex and Cursor [Subscription providers](/docs/settings/subscription-providers) run their own agent loop with their own tools, such as editing files and running commands in the chat's agent workspace folder. In Agent mode Slashspace also gives them your connectors, and all three can start sub-agents through a built-in bridge. They don't propose mode or model switches. Claude Code and Codex permission requests appear as approval rows. How often they ask depends on the permission mode or approval policy you set in Settings. A request left unanswered for 10 minutes is declined. ## What tool results share Whatever a tool returns becomes part of the conversation: it's sent to the chat's model provider (through Slashspace's hosted service for Slashspace models), saved in the chat's history, and readable by chats connected downstream. Keep that in mind with private email or files. See [Privacy and your data](/docs/privacy). ## Related - [Sub-agents](/docs/ai-features/sub-agents) - [Orchestrator mode](/docs/ai-features/orchestrator) - [Custom MCP connectors](/docs/settings/mcp-servers) # Image generation (https://www.slashspace.ai/docs/ai-features/image-generation) Generate images from a chat node in Image mode, use images on the canvas as references, and keep every version next to the conversation that made it. Visual work is part of exploring an idea: concept art, thumbnails, diagrams, a mood image for a scene you're writing. In **Image** mode, a chat node turns your prompts into pictures, and each one lands on the canvas as its own Image node, connected to the conversation that inspired it. Nothing gets overwritten, so every version and its prompt stay in view. {/* VIDEO: Switch a chat node to Image mode, send a prompt, watch the connected Image node appear, then click Edit on the result and send an edit. */} ## Switch a chat node to Image mode Choose **Image** in the mode selector at the bottom left of the composer, or press `⌘3` (`Ctrl+3` on Windows and Linux) while the chat node is active. The mode is saved per node. The model dropdown then lists image models, grouped by provider, with a "Search image models..." box. The list changes over time; a Gemini Flash image model is the default. ## Generate an image Describe what you want and send. State concrete settings when they matter, such as "16:9", "4K" or "PNG". Naming a different image model in the prompt switches the dropdown to it. Each prompt creates one image in a new Image node beside the chat. The node shows "Generating image…" with a **Cancel** button, then the image with its prompt underneath. A chip in the chat tracks the status; click it to jump to the image. You can keep sending prompts while an image generates. [Image node](/docs/nodes/image-node) lists every state the node can show. A text model reads your request first: the node's chat model (**Auto** unless you changed it). To change it, switch to **Chat**, pick a model and switch back; your image model stays selected. If the request is unclear, or asks for something the image model can't do, you get a question first, either as a chat reply or as **Before I generate** in the Image node. Click an option or type in "Or tell me what to change…". Nothing is spent on the image until you answer. ## Use reference images Connect an Image node into an Image-mode chat to edit it or use it as a reference. Uploaded, pasted and generated images all count, including ones that reach the chat through a chain or a group. A connected Scratch pad is context for the conversation, not a reference picture. A connected image is used only when your request calls for it. If several are connected, say which one you mean: "edit the fox image", "use the second image for the style". If the model can't take reference images, **Before I generate** offers one that can, or **Generate without them**. ## Iterate and branch - **Follow-up prompts** in the same chat make new images that build on the conversation. Each is a fresh generation, not an edit of the last picture. - **Edit** on a finished image creates a new Image-mode chat already connected to it, so your next prompt edits that exact picture. - **Submit in new branch** (`⌘⇧Enter`, or `Ctrl+Shift+Enter` on Windows and Linux) tries another direction in a branch that keeps Image mode and the image model. See [Branching](/docs/canvas/branching). ## Cost and requirements Hosted image generation needs an active plan: it's included with the yearly subscription and Lifetime. Other accounts see **Upgrade to generate images** when they send. Each generation is paid from your hosted wallet at the model's price. Failed and cancelled generations aren't charged. The text model that reads your prompt is billed like any other chat message. When the wallet runs out, the Image node shows **Out of usage balance**: "Your hosted wallet has insufficient funds for image generation. Add funds or use your own OpenRouter key in Settings → Providers. Your license is unchanged." That page is now **Settings > API Providers**. **Add funds** opens your account in the browser. See [Plans and billing](/docs/plans-and-billing). ### Use your own OpenRouter key 1. In **Settings > API Providers**, choose **OpenRouter** and save your API key. 2. Turn on **Use my OpenRouter key for image generation**. Generations then bill your OpenRouter account, and finished images say "via your OpenRouter key". With the switch on but no key saved, the wallet pays. The switch is greyed out on plans without it, and a key alone doesn't enable Image mode. ## Where generated images are saved Each image is saved as a file in an `images` folder inside the [canvas's folder](/docs/canvas/canvases-and-folders#where-canvases-live-on-disk). Click **Reveal in Folder** in the Image node's header to show the file, or use the copy button on the image. ## Related - [Image node](/docs/nodes/image-node) - [Connections and context](/docs/canvas/connections) - [Plans and billing](/docs/plans-and-billing) # Choosing a model (https://www.slashspace.ai/docs/ai-features/models) Pick a model for each chat, see where your AI can run, and set the defaults new chats start with. Different questions deserve different models. A quick definition doesn't need a frontier model, and a hard research question deserves more than the fastest one. On the canvas every chat node keeps its own model, so one canvas can mix a fast model for lookups, a strong reasoning model for the thread that matters, and two models answering the same prompt side by side. The models can come from Slashspace, from a Claude, ChatGPT or Cursor subscription you already pay for, from your own API keys, or from your own computer. ## Pick a model for a chat The model button sits at the right of a chat's message box, next to the microphone. 1. Click the model button. 2. Type in the "Search models..." box to filter the list, or scroll. 3. Click a model. The chat uses it from the next message, and the choice is saved with the node. The picker groups models by where they run: - **Slashspace**: **Auto** first, then Slashspace's hosted models, each with a one-line description. - Your subscription providers (Claude Code, Codex, Cursor), one group each. - Every other provider you've set up, one group each: API keys, custom providers and Ollama. Only providers you've set up appear, with the models you've turned on. **Manage AI Models** at the bottom opens Settings to add more. When your hosted balance runs low, the Slashspace group shows what's left with a shortcut to add funds or upgrade. In **Image** mode the button lists image models instead (see [Image generation](/docs/ai-features/image-generation)). **Orchestrator** mode hides it, because it runs on a model Slashspace manages. {/* VIDEO: Open a chat's model picker, search for a model, and switch the chat from Auto to a hosted model. */} ## Where AI can run | Option | Setup | What you pay | Where your messages go | | --- | --- | --- | --- | | [Slashspace](/docs/plans-and-billing) (hosted, including **Auto**) | None | Your hosted wallet, which starts with the usage included in your plan | Through Slashspace's servers to the model provider | | [Subscription providers](/docs/settings/subscription-providers): Claude Code, Codex, Cursor | Install and sign in to Claude Code or Codex once, or paste a Cursor API key | Your existing Claude, ChatGPT or Cursor plan | From a program on your computer to Anthropic, OpenAI or Cursor | | [API keys](/docs/settings/api-providers): OpenAI, Anthropic, Google, OpenRouter and more | Paste a key | The provider, per use | Straight from your computer to the provider | | [Ollama](/docs/settings/ollama) | Install Ollama and download a model | Nothing beyond your hardware | Stays on your computer (sign-in and some features still go online) | | [Custom providers](/docs/settings/custom-providers) | Add an OpenAI-compatible endpoint | Whoever runs the endpoint | To the endpoint you set | The hosted wallet pays for Slashspace's hosted models and hosted AI tools such as image generation. The other options bill their own accounts or run locally, so they work with an empty wallet. Whichever you use, chatting needs you signed in with an active plan: see [Plans and billing](/docs/plans-and-billing) for prices and [Privacy and your data](/docs/privacy) for what leaves your computer. ## Auto **Auto** "Picks the right Slashspace model for each message". Casual questions and short rewrites go to a fast, low-cost model; multi-step reasoning, coding and careful writing go to a stronger one; expert-level work and long multi-document analysis go to a frontier model. Messages in Agent mode, messages with images and very long conversations never go to the lightest models, so tools and image reading keep working. Auto chooses once per message and keeps that model for the whole reply, including every tool step. You pay for what the chosen model uses. New chats start on **Auto** unless you change your default. To get the same model every time, pin one of the hosted models listed below it. In Agent mode the AI can also suggest switching the chat's model, and nothing changes until you approve (see [Agent mode and tools](/docs/ai-features/agent-mode)). ## Reasoning effort Reasoning models can think before they answer. Reasoning effort sets how much. 1. Click **+** (tooltip "Tools") at the left of the message box. 2. Open **Reasoning** and pick **Low**, **Medium**, **High** or **Max**. You can also type `/` and choose an entry such as **Reasoning: High** (see [Slash commands, skills and prompts](/docs/ai-features/slash-commands)). The default is **Medium**. The setting belongs to the chat, and branches copy it. Reasoning effort currently reaches models you run with your own **OpenAI**, **Anthropic** or **Google** API key, and only the reasoning-capable ones act on it. For Anthropic models, **Max** behaves like **High**. Slashspace's hosted models, subscription providers, Ollama and other providers ignore it. ## Default models A chat uses the first of these that's set: 1. The model you picked in the chat itself. 2. The canvas's **Default model**. 3. **Default Chat Model** in Settings (**Auto** out of the box). Chats without their own pick follow the defaults as they change. A branch starts on its parent's model. ### Global defaults Open **Settings** > **Default Models**: - **Default Chat Model**: used when neither the chat nor its canvas sets a model. - **Title Generation Model**: names a chat after its first reply, and again when you click **Generate Title** (the sparkle beside a chat's title). As the setting says, "Pick a lightweight low cost model for this." It uses Slashspace out of the box; if you change it, choose a Slashspace model or a small model from one of your API keys. Image models aren't set here. Pick one in an Image-mode chat instead. ### Canvas default Click the gear in the canvas header to open **Canvas settings**, then choose a **Default model**. The **Global default** tag shows while the canvas follows your Default Chat Model; picking a model here stops that. It suits a canvas with one job, such as an Ollama model on a canvas of private documents. ## Tips - **Compare models on the same question.** Branch the prompt once per model, pick a different model in each branch, and send. The answers sit side by side with the same context. See [Branching](/docs/canvas/branching#compare-models). - **Use a model that can see images.** When Image nodes, a Scratch pad or other images feed a chat, the model needs vision. Auto handles this for you. Among hosted models, the descriptions flag text-only ones; for your own keys and Ollama, check the model's details with its provider. # Orchestrator mode (https://www.slashspace.ai/docs/ai-features/orchestrator) Give a chat a goal and it builds the canvas for you: chats, notes and sources, already connected. Sometimes you want the AI to lay out the work, not only answer. Give an Orchestrator chat a goal and it builds a canvas of question chats, notes and sources, wired so each chat sees what it needs. Then you explore it. Agent mode answers inside its chat and never adds or edits nodes (apart from sub-agent chats). In Orchestrator mode the canvas is the answer: it adds and connects nodes, renames them and rewrites notes, then replies in a sentence or two. {/* VIDEO: Press A, type "Plan a research canvas on the history of coffee", and watch an outline note, question chats and sources appear and connect. */} ## Start Orchestrator mode - **New chat:** press `A` on the canvas (not while typing). An Orchestrator chat appears in the middle of your view. - **Switch a chat:** pick **Orchestrator** from the mode button, or press `⌘4` (`Ctrl+4` on Windows and Linux) with the chat active. It reads the chat's earlier conversation (the recent part, if long), so "turn these ideas into nodes" works. - **Accept Agent mode's offer:** ask an Agent chat to create or connect nodes and it proposes the switch. Approve, and the chat carries on with your request. See [Mode and model switches](/docs/ai-features/agent-mode#mode-and-model-switches). ## What it can do - **Add** Chat nodes with a starting question, Text nodes, and [Web](/docs/nodes/web-node) and [Post](/docs/nodes/post-node) nodes from links. - **Connect** nodes left to right, so sources feed the chats that use them. - **Rename** any node and **rewrite** Text nodes. - **Read** the whole canvas, connected or not, and open any node in full. - **Run** a new chat's question right away, or leave it for you (hover the question, click **Rerun**). - **Delegate** research to [sub-agents](/docs/ai-features/sub-agents), then turn their answers into nodes. It can't move, resize or delete nodes, remove connections, or add Image, Document, Scratch pad or Group nodes. Press `0` (**Tidy Up**) to lay out what it built. ## Example You press `A` and type "Plan a research canvas on how the printing press changed Europe." You might get an outline note, five question chats connected from it ("How fast did literacy spread?") and a few web pages and videos feeding them, plus a reply like "Added an outline, 5 question chats and 4 sources." Then run the chats, [branch](/docs/canvas/branching) good answers, or follow up: "Add a chat comparing this to the internet." ## While it works - Changes apply immediately, with no approval step. Each shows in the chat as a row (**Create Node**, **Connect Nodes**, **Update Node**, **Subagent**). - The view doesn't move, so zoom out or click **Fit View** to see new nodes. - Web nodes arrive with the address filled in. Click **Go** on each to capture the page. - If your goal is unclear, it may ask one short question first. - **Stop** halts it and cancels its sub-agents. Follow-ups in the same chat continue the session. Switching to another mode ends it; switching back starts a new session that still reads the earlier conversation. ## Undo its changes Each change is one undo step (a node plus its connection counts as one). Click empty canvas first, since in the message box `⌘Z` undoes typing, then press `⌘Z` (`Ctrl+Z` on Windows and Linux) per step. See [Undo and redo](/docs/canvas/organizing#undo-and-redo) for limits. ## What's different in this mode - **No model picker:** it runs on a model Slashspace chooses, not your keys, subscriptions or local models. - **No connectors, personas or agent workspace.** The reasoning level has no effect either. - **Online only:** it needs an internet connection and a signed-in account. - The chats and sub-agents it starts are ordinary chats on your models, billed like any chat ([Plans and billing](/docs/plans-and-billing)). ## Privacy Orchestrator mode runs in Slashspace's cloud. Each message sends your request, a summary of the canvas (node titles, short previews of their content, file names and locations, connections), any node it opens in full, sub-agent answers, and whatever the chat said before you switched it. Choosing the mode yourself is the go-ahead; there's no separate prompt. Agent proposals wait for approval unless **Auto-switch into Orchestrator mode** is on (**Always allow** on a proposal turns it on). Turn it off in **Canvas settings** (the gear in the canvas header) under **Orchestrator**; it covers every canvas. See [Privacy and your data](/docs/privacy). ## If something goes wrong Errors show in a banner in the chat, with **Retry** when the problem is temporary. | Message | What to do | |---------|------------| | "Canvas Orchestrator is busy", "temporarily unavailable" or "took too long to respond" | Wait, then **Retry**. | | "Couldn't connect to Canvas Orchestrator" | Check your internet connection. | | "This Orchestrator session has ended" | Switch to **Chat**, back to **Orchestrator**, and resend. | | "Sign in to Slashspace again" | **Sign out** in Settings > **Account**, then sign back in. | | "There's too much context for Orchestrator" | Start a fresh Orchestrator chat (`A`) with a shorter request. | | Anything that keeps happening | Email support@slashspace.ai. | ## Related - [Agent mode and tools](/docs/ai-features/agent-mode) - [Sub-agents](/docs/ai-features/sub-agents) - [Connections and context](/docs/canvas/connections) # Personas (https://www.slashspace.ai/docs/ai-features/personas) Save a system prompt as a persona and put it on any chat node, so each branch can read the same material through a different lens. A persona is a saved system prompt: standing instructions that shape how the AI answers. A devil's advocate who attacks every claim, an editor who tightens prose, a tutor who explains step by step. You write it once and put it on any chat node. Each chat node can have its own persona, so branches of one conversation can look at the same material through different lenses. {/* VIDEO: Create a persona from Tools > Personalities > Add New, check it on a chat, then press ⌘B and give the new branch a different persona. */} ## Create and manage personas 1. Open **Settings** and click **Persona Prompts** (under **Templates**). 2. Click **Create Persona**. 3. Fill in the form: - **Name**: how the persona appears in menus. - **Color**: a swatch that marks its card on the Persona Prompts page. - **Prompt**: the instructions the AI follows, such as what role to play, what to look for and how to shape the answer. 4. Click **Create**. You can also create one from a chat: click **+** in the message box, open **Personalities** and choose **Add New**. The new persona isn't applied until you check it. To edit a persona, click its card on the Persona Prompts page, change any field and click **Update**. Chats that use it follow the new version from their next message. To delete it, open the card, click **Delete**, then **Yes, Delete**. > **Note:** Turn a persona off in the chats that use it before you delete it. A chat whose persona was deleted fails to send until you pick a different persona there. ## Apply a persona to a chat - **Tools menu**: click **+** (tooltip "Tools") in the message box, open **Personalities** and check a persona. - **Slash menu**: type `/`, open **Personas** and pick one, or type part of its name. See [Slash commands, skills and prompts](/docs/ai-features/slash-commands). A chat holds one persona at a time; picking another replaces it. The active one has a check mark in both menus (the `/` menu adds "Active, select to remove"). The chat node itself doesn't show it, so open either menu to check. To turn a persona off, pick it again. The persona applies from your next message, and replies already in the chat stay as they are. It affects only this chat: other chats, even connected ones, keep their own. Branches you fork from a chat start with the parent's persona, and changing it in the branch leaves the parent alone. The exception is a chat dragged out of a handle, which starts with none. See [Branching](/docs/canvas/branching). Personas work in every mode except [Orchestrator](/docs/ai-features/orchestrator). The menus still let you check one there, but it has no effect in that mode. ## Example: two lenses on one question Say a chat about putting your newsletter behind a paywall is connected to your reader survey and two competitors' pricing pages. 1. Create a "Devil's advocate" persona ("Argue against the plan. Find its weakest assumptions and the strongest counterevidence.") and an "Operator" persona ("Turn the plan into steps, costs and risks, and name the first thing to measure."). 2. In the chat's message box, press `⌘B` (`Ctrl+B` on Windows and Linux) twice to make two empty branches. 3. Put **Devil's advocate** on one branch and **Operator** on the other, then send both "Should I do this?" Both branches read the same conversation and sources through their parent, so only the persona differs. ## Where personas are stored Personas live on this computer, in Slashspace's app data folder rather than with your canvases (on a Mac, `~/Library/Application Support/Slashspace/config/persona-prompts.json`). They don't sync between devices. A chat node records only which persona it uses, so a copied canvas arrives without its personas. A persona's text is sent to the model with each message, like the rest of the conversation. # Slash commands, skills and prompts (https://www.slashspace.ai/docs/ai-features/slash-commands) Type / in any chat to add a skill or a saved prompt to your message, pick a persona, switch connectors or set reasoning effort. The instructions you reuse, like a critique rubric, a summary format or a research method, shouldn't be retyped in every chat. Save them once as files, and `/` puts them one keystroke away on any canvas. The same menu also sets up the chat itself: persona, connectors and reasoning effort. {/* VIDEO: Type / in a chat node, drill into Skills, pick a skill, then add a saved prompt and send. */} ## Open the menu Type `/` in a [chat node's](/docs/nodes/chat-node) composer or the [side view](/docs/canvas/side-view), at the start of a line or after a space. **Skills**, **Personas** and **Connectors** appear as sections with a count (empty ones are hidden), followed by the reasoning options and your prompts. | Key | What it does | |-----|--------------| | `↑` `↓` | Move through the list | | `→` | Open the highlighted section | | `Enter` or `Tab` | Select the item, or open the section | | `←` or `Esc` | Back out of a section (or click its header) | | `Esc` at the top level | Close the menu | Keep typing after `/` to filter. At the top level the search covers every section; inside a section it covers only that section. Skills and prompts match on name, description or file path, and `reasoning`, `persona` or `connector` lists everything of that kind. A space closes the menu, so search with one word. ## What each item does | Item | When you select it | |------|--------------------| | Skill or prompt | Inserts a chip. When you send, the file's full text goes to the model with your message. | | Persona | Applies that [persona](/docs/ai-features/personas) to this chat. The active one reads "Active, select to remove". | | Connector | Turns that connector on or off for this chat only. **Manage connectors** opens Settings > **Connectors**. | | **Reasoning: Low**, **Medium**, **High**, **Max** | Sets how much the model reasons before answering, on [models that support it](/docs/ai-features/models). Medium is the default. | Personas, connectors and reasoning apply right away and leave no chip. They're the same settings as the **+** (**Tools**) menu in the composer. ## Prompts Prompts are Markdown files in `~/SlashspaceOS/.prompts`, next to your canvases (`~` is your home folder, `C:\Users\` on Windows). Subfolders work too. 1. Create the `.prompts` folder if it isn't there. Finder hides names that start with a dot: press `⌘⇧.` to show them. 2. Save a `.md` file in it, for example `critique.md`. 3. Type `/crit` in any chat. New files appear right away, no restart needed. An optional header sets the prompt's menu name and subtitle: ```md --- name: Critique description: Score a draft against my rubric --- Critique the draft below. Rate clarity, argument and evidence from 1 to 5, quote the weakest sentence in each area, and suggest one fix for each. ``` Without `name`, the menu shows the file name; without `description`, the file path. Other header fields are ignored, and the header is never sent. To use it, insert `/Critique`, then paste your draft after the chip. Your message shows only the chip, but the model reads the full prompt. The file is read each time the chat sends, so edits apply from your next message. ## Skills A skill is a folder containing a file named `SKILL.md` that teaches the AI a workflow. The menu lists every skill in `~/.agents/skills`, the shared folder the skills.sh installer uses, so skills you installed that way for other AI tools show up here too. Install more from [Agent Skills](/docs/settings/skills) in Settings. Skills work like prompts: pick one to insert a chip, and its instructions go with your message. Only `SKILL.md` is sent, not other files in the folder such as scripts or examples. The menu name comes from the skill's `name` field, or its folder name. ## Which modes use what | | Agent | Chat | Image | Orchestrator | |---|---|---|---|---| | Skills and prompts | Yes | Yes | Yes | No | | Personas | Yes | Yes | Yes | No | | Connectors | Yes | No | No | No | Reasoning effort works in every mode except Orchestrator. Outside [Agent mode](/docs/ai-features/agent-mode), the Connectors section shows "Available in Agent mode". [Orchestrator mode](/docs/ai-features/orchestrator) runs in Slashspace's cloud on a fixed model: the menu still opens, but nothing in it changes what Orchestrator does. ## If something goes wrong - **"Missing slash command reference"**: the file behind a chip was moved, renamed or deleted. Remove the chip and add it again. Editing the `name` field is safe; renaming the file or folder is not. - **A chat stops sending after you delete a file:** an earlier message in that chat, or in a chat that feeds into it, still uses it. Put the file back. - **"Too large"**: one skill or prompt can hold up to 120,000 characters, and one message can expand to 400,000. - **Missing from the menu:** type part of the name, since long lists get cut short. Symlinked files and folders are skipped. # Sub-agents (https://www.slashspace.ai/docs/ai-features/sub-agents) Let the AI split a big question into parallel tasks that each run in their own chat node and report back. Big questions are often several smaller ones. Sub-agents let the AI hand each part to its own chat node, run them side by side, and bring the answers back to the chat that asked. Nothing is buried in one long reply: every sub-agent is a node you can read, check and keep talking to. {/* VIDEO: Ask an Agent-mode chat to research three topics in parallel; show the sub-agent nodes appearing to its right, a live Subagent row in the parent, and the combined answer. */} ## When the AI uses sub-agents Sub-agents are part of **Agent** mode, the default. Chat and Image modes don't use tools, so they never start them. [Orchestrator mode](/docs/ai-features/orchestrator) uses them too, for research it then turns into nodes. Ask in plain words: "research these three companies in parallel". The AI can also split a task on its own when the parts are independent. > **Tip:** If you want nodes on the canvas as the result (say, one per topic), Agent mode proposes switching to Orchestrator mode instead. See [Agent mode and tools](/docs/ai-features/agent-mode). ## What you see 1. A chat node appears for each task, connected from the parent and stacked in a column to its right. Its first message is the task the AI wrote. Being connected, it also reads the parent's conversation, like any [connected chat](/docs/canvas/connections). 2. In the parent, each sub-agent is a **Subagent** row (several are grouped under a heading like "Ran 3 subagents"). Open a row to follow it live: its status, the tool it's using, the latest lines of its reply, **Jump to node** and **Cancel**. 3. Each sub-agent reports its final answer back. Once all have, the parent carries on, usually by combining them. ## Choosing models Sub-agents use the parent chat's model unless you name others: "have three sub-agents argue this, one on Claude, one on GPT and one on Gemini." Loose names work. The AI can pick **Auto**, any Slashspace hosted model, or any chat model in your [model picker](/docs/ai-features/models). If a name matches nothing, that sub-agent fails before a node is created and the parent is told which models it can use. Each sub-agent is billed like any chat on its model: hosted models (including **Auto**) draw from your hosted wallet; your own API keys, subscription providers and Ollama don't. Since each one reads the parent's conversation, a long parent thread makes every request bigger. See [Plans and billing](/docs/plans-and-billing). ## Limits | Limit | Details | |-------|---------| | Depth | One level. Sub-agents can't start their own, even when you chat with them later. | | Time | 10 minutes each, then it's stopped and reported as timed out. | | How many | No fixed number. They all run at once. | | Stay on the canvas | Switching canvases or quitting mid-run marks the run interrupted. Ask again if you still need it. | If a sub-agent's tool needs approval, the card appears on its own node while its row says "Subagent is waiting to continue…". Approve it there before the time runs out. ## Stopping and retrying **Cancel** on a row stops that sub-agent, as does stopping the chat on its node. Cancelled runs don't wake the parent, so if you cancel them all, send the parent a message to carry on. When a sub-agent fails or times out, the parent is told why and can retry it in the same node, on the same model or a different one. You can ask too: "retry the one that failed on another model." ## Claude Code, Codex and Cursor [Subscription providers](/docs/settings/subscription-providers) run tools inside their own agent, so Slashspace adds its sub-agent tool there and asks the agent to use it instead of built-in helpers that run out of sight. Each sub-agent is still a connected chat node with the same limits, and its answer goes back into the session. The parent shows a **Subagent** row without the live preview, so watch the child node. **Stop** on the parent cancels its sub-agents. ## Continuing a sub-agent's chat A sub-agent node stays on the canvas as a normal chat. Open it to see the whole run, tools included, then keep talking to it, [branch](/docs/canvas/branching) from it, or delete it. After your next message in the parent, earlier sub-agent answers shrink to short excerpts there. The full answer stays on the sub-agent's node, so connect that node to any chat that needs all of it. ## Related - [Agent mode and tools](/docs/ai-features/agent-mode) - [Orchestrator mode](/docs/ai-features/orchestrator) - [Choosing a model](/docs/ai-features/models) # Voice input (https://www.slashspace.ai/docs/ai-features/voice-input) Dictate messages into any chat, choose which microphone Slashspace listens to, and fix dictation when it doesn't work. Thinking out loud is often faster than typing, especially when you're reacting to a source or talking through a half-formed idea. Voice input turns what you say into text in the message box, so you can ramble first and tidy up before you send. Dictation works in every chat composer: in a [Chat node](/docs/nodes/chat-node) on the canvas and in the [side view](/docs/canvas/side-view). It isn't available in Text nodes. ## Dictate a message 1. Click the microphone button in the composer toolbar, next to the model button. Or use a shortcut from the table below. 2. Speak. A timer and a live waveform show while Slashspace listens. 3. Press `Enter` or click the check mark. After a moment of "Transcribing...", the text appears in the message box. 4. Edit the text if you need to, then send it as usual. Nothing reaches the AI until you send it. | Action | Shortcut | |--------|----------| | Start dictating (message box empty) | `Space` twice, quickly | | Start dictating (cursor in the message box) | `Ctrl+M` (the Control key on a Mac too, not Command) | | Finish and transcribe | `Enter` | | Cancel and discard the recording | `Esc` | > **Note:** The transcript replaces whatever is already in the message box, so dictate first and type afterward. {/* VIDEO: Press Space twice in an empty chat, speak a sentence, press Enter, and show the transcript landing in the message box before sending. */} ## Choosing a microphone Slashspace uses your system's default input unless you pick another. Two places set the same preference, and it applies to every chat: - **From the composer:** hover the microphone button and click the small arrow beside it. In the **Microphone** menu, pick **System Default** or a specific device. - **From Settings:** open **Settings > System Preferences** and choose a **Preferred Microphone** under **Audio Settings**. If the microphone you picked is unplugged, dictation won't start until you switch back to **System Default**. ## How transcription works Transcription uses Groq's Whisper speech-to-text model, so it needs an internet connection. The recording is sent only when you finish it; a cancelled recording never leaves your computer. - **Without a Groq key** (the default), your audio goes through Slashspace's server to Groq. There's nothing to set up, and it doesn't draw from your hosted wallet. - **With your own Groq key** saved in [API providers](/docs/settings/api-providers), Slashspace sends the audio straight to Groq, and the usage counts against your Groq account. See [Privacy and your data](/docs/privacy) for what else leaves your computer. ## Permission and troubleshooting On a Mac, the first time you dictate (or open the microphone list), macOS asks whether Slashspace can use the microphone. Allow it. If the recording closes right away, or no text appears after "Transcribing...", Slashspace doesn't show an error. Check: - **Mac:** System Settings > Privacy & Security > Microphone has Slashspace turned on. If you declined the prompt earlier, macOS won't ask again. - **Windows:** your privacy settings let desktop apps use the microphone. - **Microphone:** **System Default** is selected in the **Microphone** menu. - **Internet:** you're online. - **Groq key:** if you added one, it's still valid in API providers. Removing it switches back to the built-in transcription. Still stuck? See [Troubleshooting and support](/docs/troubleshooting). # Branching (https://www.slashspace.ai/docs/canvas/branching) Fork any conversation into a new branch to chase a tangent, compare answers, or try another model without touching the original thread. Exploration is full of tangents. In a chat app, following one derails the thread you care about. On the canvas you branch instead: a branch is a new chat node connected to the one you came from, so it starts with everything the parent knew, and the original thread stays as it was. Branches sit side by side, so you can follow several ideas from the same point, compare the answers, and see weeks later where each line of thinking split off. {/* VIDEO: Type a follow-up in a chat, press ⌘⇧Enter to send it into a new "Split" branch, then select a sentence in the answer and ask about it. */} ## Fork a conversation - **Submit in new branch**: type your message, then click the split-arrows button next to **Send** (tooltip "Submit in new branch") or press `⌘⇧Enter` (`Ctrl+Shift+Enter` on Windows and Linux). A chat titled "Split" appears to the right and runs your prompt right away. The original chat never receives the message. - **Empty branch from the composer**: press `⌘B` (`Ctrl+B` on Windows and Linux) while typing. An empty chat titled "New Branch" appears to the right and the view moves to it. Your draft stays where it was. - **Empty branch from the canvas**: click a node to make it active, then press `⌘B`. The new "New Branch" chat becomes the active node and the view fits to it. On Windows and Linux, double-click the node's right handle instead. ## Branch from any node Every node has a right handle, so a note, web page, document or post can start conversations too. Branch one source three times, give each chat its own question, and the threads never bleed into each other. - **Drag** from the right handle and drop on empty canvas. A connected chat appears where you dropped it. - **Double-click** the right handle (hovering it shows a "create branch" pill). A connected chat appears to the right and becomes active. From an Image node or a Scratch pad, both gestures open a picker: **Chat**, **Image** or **Scratch pad**. ## Ask about a passage To dig into one sentence of an answer rather than continue the whole thread, select text in an AI response. A bubble appears with the placeholder "Ask about this… (↵ to expand)". - Type a question and press `Enter`. A branch titled with the start of your selection opens and sends `Regarding: "…"` with the quoted passage, followed by your question. - Press `Enter` without typing to ask the AI to expand on the passage. This is a focused follow-up, not a fork: its first message quotes the passage, so the answer centers on that text. The branch is still connected to the chat, so the AI has the full conversation for reference. ## What a branch sees A branch is joined to its parent by an ordinary connection, so it reads the parent's content (for a chat, its whole conversation) plus everything connected upstream of the parent. Sibling branches can't see each other, and nothing flows back into the parent. See [Connections and context](/docs/canvas/connections). The connection is live: if you keep talking in the parent after forking, the branch sees those messages the next time you send in it. Branches made with **Submit in new branch**, `⌘B`, a double-click or the selection bubble also copy the parent chat's setup: model, mode, persona, reasoning level, connectors and context mode. Changing them in the branch leaves the parent alone. A chat dragged out of a handle keeps the model and connectors but starts in **Agent** mode. > **Note:** If the parent uses **Isolated context**, so does the branch, and an isolated chat ignores its incoming connections, including the one from its parent. Switch the parent to **Summarized context** before branching if the branch should see it. ### Compare models Branches that share a parent answer from the same context, which makes them a fair way to compare models. 1. In the parent's composer, press `⌘B` once per model you want to try. 2. In each branch, pick a model from the model dropdown ([Choosing a model](/docs/ai-features/models)). 3. Send the same prompt in each. The answers sit side by side. ## Working with many branches - **Name them.** Branches keep their starting title ("Split", "New Branch" or the quoted passage); a chat that starts untitled is titled automatically after its first reply. Click a title to rename it, or hover it and click **Generate Title**. - **Color-code** a line of thinking with **Change Color** in the node header. - **Group** related branches. A branch of a node inside a [group](/docs/nodes/group-node) lands in the same group. - **Tidy Up** (`0`) lays the canvas out left to right, each branch to the right of its parent. - **Collapse** finished branches to their last exchange. - **Zoom out.** Below 30% zoom each node shows its title as a pill; click one to fly back to it. More in [Organizing the canvas](/docs/canvas/organizing) and [Navigating the canvas](/docs/canvas). # Canvases and folders (https://www.slashspace.ai/docs/canvas/canvases-and-folders) Create, organize and find your canvases in the sidebar, and see where each one is stored on your computer. Exploratory work rarely fits on one canvas. A project grows into a canvas per question, source batch or draft, and you return to them over weeks. Folders keep each project together. Every canvas is also a plain folder on your disk: ordinary files that you own and can back up. ## The sidebar The sidebar on the left lists your folders and canvases. It calls canvases spaces, which is why its button says **New Space**. The top holds **New Folder**, **New Space** and a **Search** field. The bottom holds **Settings** and the theme toggle. To hide or show it, click the panel button at the top left of the canvas (**Hide Sidebar** or **Show Sidebar**). Drag its right edge to resize it. ## Create canvases and folders Click **New Space** to create a canvas at the top level and open it. To create one inside a folder, hover the folder and click its **+** button, or choose **New Space** from the folder's **⋮** menu. A new canvas is named "untitled" ("untitled 2" if that's taken) and holds one empty chat node. Click **New Folder** to add a folder named "New Folder" at the top level, then rename it. ## Rename, duplicate or delete Hover a canvas or folder and click **⋮** to open its menu. - **Rename**: type the new name and press `Enter` (`Esc` cancels). Double-clicking a name does the same. - **Duplicate** (canvases only): makes a full copy, with every node, connection and chat, in the same folder. A copy of "Climate" is named "Climate-copy". - **Delete**: see below. Deleting a canvas asks you to confirm (**Delete Space?**, then **Yes, Delete**) and removes its folder from your disk. It doesn't go to the Trash and can't be undone, so [export](/docs/canvas/export) anything you might want later. A folder can only be deleted once no canvases are left in it or its subfolders. Otherwise Slashspace tells you how many it holds, so you can move or delete them first. An empty folder is deleted right away, with no confirmation, along with any other files you put in it on disk. ## Move canvases and folders Drag a canvas or folder onto a folder to move it inside; the folder highlights when you can drop. Drop on empty space in the list to move it back to the top level. Nest folders the same way, by dragging one onto another. {/* VIDEO: Dragging a canvas onto a folder in the sidebar, then dragging that folder into another folder. */} If the destination already has a canvas with that name, the one you move gets a number added. Folders are listed first, A to Z. Top-level canvases are sorted by last activity, most recent first. ## Search the sidebar Type in **Search** to filter canvases by name. Folder names match too, so typing a folder's name lists every canvas in it. Press `Esc` to clear. To find something inside a canvas, use [Search](/docs/canvas/search) instead. ## Canvas limit on older plans Some older plans cap how many canvases you can have. On those, the sidebar shows a counter such as **2 of 3 spaces**, and duplicates count toward the cap. Current plans include unlimited canvases: see [Plans and billing](/docs/plans-and-billing). ## The canvas header The bar across the top of the canvas shows the folder path and canvas name, such as `Research / Climate / Sources`. It also holds the sidebar toggle, [Canvas settings](/docs/canvas) and the [side view](/docs/canvas/side-view) toggle. ## Where canvases live on disk Everything lives in a `SlashspaceOS` folder in your home folder (`~/SlashspaceOS` on Mac, `C:\Users\you\SlashspaceOS` on Windows). Each sidebar folder is a real folder, and each canvas is a folder with the canvas's name. ``` SlashspaceOS/ ├── Research/ │ └── Climate/ │ └── Sources/ │ ├── canvas.json │ ├── Context.md │ └── images/ └── images/ ``` | Item | What it holds | | --- | --- | | `canvas.json` | The whole canvas: nodes, connections, chats and layout. | | `Context.md` | A readable summary: node titles, short previews and how nodes connect. Slashspace rewrites it whenever the canvas saves, so don't edit it. | | `images` in a canvas | Images generated in that canvas. | | `images` at the top | Images you paste. | Images and documents you add from elsewhere on your computer aren't copied. The canvas links to the original, so moving or deleting it can break the node. Renaming or moving a canvas or folder in the app renames or moves its folder on disk. Characters not allowed in file names, such as `/` and `:`, become hyphens in canvas names. Don't name a folder `files` or `images`: the sidebar hides those. Slashspace doesn't watch these files. An open canvas keeps saving over its `canvas.json`, so edits you make on disk can be lost. Quit Slashspace before changing anything by hand. To back up, copy the whole `SlashspaceOS` folder so pasted images come along, and keep the originals of files you added. To share one canvas as Markdown or JSON, use [Export](/docs/canvas/export). For what leaves your computer, see [Privacy and your data](/docs/privacy). # Connections and context (https://www.slashspace.ai/docs/canvas/connections) Control exactly what each chat reads by connecting nodes, choosing a context mode, and wiring in sources with mentions, the navigator and groups. In long-context work, the hard part is controlling what the model reads when it answers. On the canvas, connections are that control. A connected chat reads what flows into it, so a synthesis chat can read three sources while the side question next to it reads none of them. It also keeps threads from polluting each other: connect a lecture PDF to three chats, one per exam topic, and none of them sees the others' conversations. {/* VIDEO: Drag a Web node's right handle onto a chat's left handle, send a question, and watch the connection light up while the answer streams. */} ## Handles Every node has a handle on its right edge, and chat nodes also have one on their left. The right handle gives context; the left handle takes it. (Image nodes have a left handle too, but it only accepts other Image nodes.) Handles appear when you hover a node, and connected ones turn orange. - **Connect**: drag from a node's right handle onto a chat node's left handle. Dropping on empty canvas instead creates a new, already connected chat (see [Adding nodes](/docs/canvas/creating-nodes)). - **Remove**: click the connection line to select it, then press `Backspace` or `Delete`. The toast that appears offers **Restore**. - **Move**: drag either end of a connection onto a different handle. ## How context flows Context flows left to right and follows chains. If A connects to B and B connects to C, the chat in C reads both A and B. A node that reaches a chat by more than one path counts once. Context is gathered each time you send, so an edit to a connected note is picked up by your next message. A branch is a new chat connected to the one you forked, so it reads the whole parent conversation. See [Branching](/docs/canvas/branching). | Node | What a connected chat reads | | --- | --- | | Chat node | Its full conversation | | Text node | Its text | | Web node | The captured page text | | Post node | The transcript, when one is available | | Document node | The document's text. A PDF is sent as the file itself until its conversion finishes. | | Image node | The image itself | | Scratch pad | Your drawing, as an image. An empty pad adds nothing. | | Group | Every node inside it (see [Groups as context](#groups-as-context)) | In Image mode, connected Image nodes also become candidate reference images. See [Image generation](/docs/ai-features/image-generation). ### Large inputs Nothing is summarized or trimmed: every connected node is sent in full with every message. Enough big sources can exceed a model's context window, and the request fails; disconnect what the chat doesn't need, or pick a model with a larger window. Document text larger than 512 KB is left out (the AI is told it was skipped). An Office file that's still converting is skipped with a notice such as "'Report.docx' is still being processed and wasn't included in context." > **Note:** RAG mode, which sent only matching excerpts of connected content, has been retired. Connected content is now sent to the model directly. ## Context modes Every chat node has a context mode that decides what it reads beyond its own conversation. - **Summarized context** (the default). A chat with no connections gets a short summary of every other node on the canvas: titles, previews, each chat's latest messages, and how nodes connect. In Agent mode, the AI can open the full content of any node that looks relevant. Once you connect something, the chat reads its connections in full instead. - **Isolated context**. The chat reads only its own conversation: no canvas summary, no connections. Use it for a question unrelated to the canvas, or a critique you don't want anchored by your notes. To switch, hover the chat's left handle and click the icon inside it (the lightning icon is **Summarized context**; the tooltip names the current mode). You can't switch while the chat is generating. The switch only appears while the chat has no incoming connections. Once something is connected, the handle shows the number of direct connections instead, such as "2 incoming connections". > **Note:** An isolated chat keeps ignoring connections you add later, and a branch copies its parent's mode. If a chat's left handle stays gray after you connect something, it's still isolated: remove its incoming connections, switch it to **Summarized context**, then reconnect. ## Mentions create connections Type `@` in a chat's message box and pick a node from the "Mention a node" menu. The chat connects to that node right away, as if you had dragged the connection, and the connection stays after you send. Deleting the mention from your message doesn't disconnect the node; remove the connection on the canvas. Groups show their size in the menu, such as "group • 3 children", and mentioning one connects the whole group. See [Chat node](/docs/nodes/chat-node). ## Add context from the navigator The node navigator, the column of short lines on the left edge of the canvas, lets you wire in sources that are far away. Click a chat to make it active, then drag another node's line to the right ("Drag right to add as context"). Connected lines turn orange; drag one right again to disconnect it ("Drag right to remove context"). Groups are listed at the bottom with their child count. Click or drag a group's line to add or remove the whole group. See [Navigating the canvas](/docs/canvas). {/* VIDEO: With a chat active, drag two lines in the left-edge navigator to the right to add them as context, then drag one again to remove it. */} ## Groups as context Drag from a group's right handle to a chat, or mention the group, and the chat reads everything inside it. - Each node inside contributes as it would on its own, in reading order: top to bottom, then left to right. - It stays live: nodes you add to or move out of the group are picked up by your next message. - Only nodes inside the group count. Sources connected into a grouped node from outside aren't included. - A node that arrives both directly and through the group counts once, and a chat inside the group never reads itself. ## See what a chat reads Click a chat node to highlight every connection feeding it in orange, all the way up the chain. While the chat generates, light travels along those same connections. ## Edge styles Choose how connection lines look under **Edges** in **Canvas settings** (the gear icon in the canvas header): **Default**, **Straight**, **Step**, **Smooth step** or **Simple bezier**. The style is purely visual and applies to every canvas. ## Related - [Branching](/docs/canvas/branching) - [Group](/docs/nodes/group-node) - [Agent mode and tools](/docs/ai-features/agent-mode) # Adding nodes (https://www.slashspace.ai/docs/canvas/creating-nodes) Every way to put a question, note, link, file or sketch on the canvas, from one-key hotkeys to paste and drag-and-drop. When a question comes to you mid-read, or you find a source worth keeping, you want it on the canvas in a keystroke, next to the work it belongs with, before the thought slips away. Use whichever of these methods is closest to hand. ## The toolbar and its hotkeys The toolbar sits at the bottom center of the canvas. Click a node type or press its key, then click where you want the node. While you place it, the cursor is a crosshair and the toolbar reads "Click anywhere on the canvas to add" plus the node type. Press `Esc`, or the same key again, to cancel. | Node | Key | Use it for | | --- | --- | --- | | [Chat node](/docs/nodes/chat-node) | `C` | Asking the AI a question | | [Text node](/docs/nodes/text-node) | `T` | Your own notes | | [Image node](/docs/nodes/image-node) | `I` | A picture from your computer (**Add From Computer**) | | [Scratch pad](/docs/nodes/scratch-pad-node) | `S` | A freehand sketch | | [Post node](/docs/nodes/post-node) | `V` | A YouTube, TikTok, Instagram, Facebook, X or LinkedIn post | | [Document node](/docs/nodes/document-node) | `D` | PDFs, Office and text files (a file picker opens when you click the canvas) | | [Web node](/docs/nodes/web-node) | `W` | A web page (the toolbar calls it **Web page node**) | | [Orchestrator](/docs/ai-features/orchestrator) chat | `A` | A goal for the AI to build out on the canvas. It has no button: the chat opens mid-view, ready to type in. | The toolbar also holds **Tidy Up** (`0`, see [Organizing the canvas](/docs/canvas/organizing)) and **Search** (`⌘K`, or `Ctrl+K` on Windows and Linux, see [Search](/docs/canvas/search)). > **Note:** The single-key hotkeys don't fire while you're typing in a chat composer, a Text node or any other text field. Click an empty spot on the canvas first. ## Cmd-click for a quick chat Hold `⌘` (`Ctrl` on Windows and Linux) and click any empty spot. A new chat node appears right there, next to whatever prompted the question. An empty canvas reminds you with "Use the Toolbar or Cmd/Ctrl + Click anywhere to create a node". ## The right-click menu Right-click an empty spot. Under **Add**, choose **Chat Node**, **Text Node**, **Web Node**, **Post Node**, **Document Node** or **Image Node**. The last two are labeled **Beta**. The node appears where you right-clicked (**Document Node** opens a file picker first). Scratch pads aren't in this menu: use the toolbar or press `S`. ## Drag from a node's handle Drag from the handle on a node's right edge and release on empty canvas. You get a new chat node wired to the node you started from, so the chat sees it as context: the natural way to ask about a web page, a document or an earlier answer. See [Connections and context](/docs/canvas/connections). From an Image node or a Scratch pad, a **Create node** menu opens instead, offering a connected **Chat**, **Image** or **Scratch pad**. Double-clicking the right handle does the same without dragging (see [Branching](/docs/canvas/branching)). New nodes only come out of the right handle: dragging from a left handle onto empty canvas shows "Nodes can only be created from the right handle". {/* VIDEO: Drag from a Web node's right handle onto empty canvas, release, and type a question into the connected chat that appears. */} ## Paste Click the canvas and press `⌘V` (`Ctrl+V` on Windows and Linux). Slashspace picks the most fitting node: | You paste | You get | | --- | --- | | Images or other files | Image or Document nodes, by the drag-and-drop rules below | | Nodes you copied | Copies of those nodes (see [Organizing the canvas](/docs/canvas/organizing)) | | A YouTube, TikTok, Instagram, Facebook, X or LinkedIn post link | A Post node | | Any other link | A Web node | | Plain text | A Text node | Pasted content lands at your mouse pointer, or mid-view if the mouse hasn't been over the canvas yet. As with hotkeys, pasting while you're typing in a text field puts the content there instead. ## Drag files in from your computer Drag files from Finder or File Explorer onto the canvas. Each file becomes a node where you drop it, and several files line up in a row. - Images (PNG, JPG, GIF, WebP, SVG) become [Image nodes](/docs/nodes/image-node). - PDFs, Word, PowerPoint and Excel files, plus text and code files such as Markdown, CSV and JSON, become [Document nodes](/docs/nodes/document-node). - Anything else shows a message such as "File type .zip is not supported". ## Nodes the AI adds for you - **Branches.** **Submit in new branch** forks a conversation into a new connected chat node ([Branching](/docs/canvas/branching)). - **Sub-agents.** In Agent mode, the AI can hand parts of a task to new connected chat nodes ([Sub-agents](/docs/ai-features/sub-agents)). - **Generated images.** Each image from Image mode arrives as a new connected Image node ([Image generation](/docs/ai-features/image-generation)). - **Orchestrator.** Give an Orchestrator chat a goal and it creates chat, text, web and post nodes and wires them together ([Orchestrator mode](/docs/ai-features/orchestrator)). # Export (https://www.slashspace.ai/docs/canvas/export) Take your work out of a canvas: download the whole thing as Markdown or JSON, or copy and open individual pieces. Exploration usually ends in something you take elsewhere: a draft, notes, a backup. So a canvas is never a dead end: download the whole thing as one file, or pull out single pieces (a reply, an image, a converted document) as you go. ## Download a whole canvas 1. Right-click an empty spot on the canvas. 2. Hover **Download Canvas**. 3. Choose **.md** for Markdown or **.json** for JSON, then pick where to save it. The file is named after the canvas. {/* VIDEO: Right-click the canvas, choose Download Canvas > .md, then open the file in a notes app. */} ### Markdown The Markdown file is for reading or pasting into a document or notes app. It contains: - The canvas name as the top heading. - One section per node, headed by the node's title, in roughly the order you added the nodes (not their position on the canvas). - For each chat node, every message in the conversation under a `user` or `assistant` heading. Only the text of chat messages is included. Tool calls, reasoning and images are left out, and other nodes (text notes, web pages, documents, images) appear as a title heading without their content. To take a note's text with you, copy it directly (see below). > **Tip:** Drop the `.md` file onto another canvas to bring those conversations in as a Document node, ready to connect as context. ### JSON The JSON file is a complete copy of the canvas as Slashspace saves it: every node with its full content, the connections between them, positions and sizes, and the canvas's settings. Use it as a backup or to process your canvas with your own scripts. There's no way to import a JSON file back into Slashspace as a canvas. If you drop one onto a canvas, it's added as a Document node like any other text file. ## Get individual pieces out - **A message:** hover a message in a chat node and click the **Copy** icon. You get the message text with its Markdown formatting. Code blocks, tables and diagrams in a reply have their own copy and download buttons (tables as CSV or Markdown). - **Any text:** select text in a node or in the [side view](/docs/canvas/side-view) and press `⌘C` (`Ctrl+C` on Windows and Linux). - **An image:** hover an Image node and click **Reveal in Folder** to show the image file in Finder (File Explorer on Windows). - **A document:** once a Document node shows **Ready**, click **.md** (Open Markdown) or **.pdf** (Open PDF) to open the converted version in your browser. **Reveal in Folder** in the node's header shows the file on your computer. See [Document node](/docs/nodes/document-node). Every canvas is already a folder on your computer, so you can also back up or open its files directly. See [Canvases and folders](/docs/canvas/canvases-and-folders). # Navigating the canvas (https://www.slashspace.ai/docs/canvas) Pan and zoom around a large canvas, jump straight to any node, and choose how the canvas responds to your touchpad or mouse. A big exploratory canvas only works if moving around it costs you nothing. Zoom out to see the shape of your thinking, jump to the node you need, then zoom back in to read and write. When you open a canvas, the view frames all of its nodes. ## Moving around What scrolling does depends on the **Movement** setting in [Canvas settings](#canvas-settings). **Touchpad** is the default. | Action | Touchpad | Mouse | | --- | --- | --- | | Scroll (two fingers or the wheel) | Pans | Zooms | | Pinch on a trackpad | Zooms | Zooms | | `⌘` + scroll (`Ctrl` + scroll on Windows and Linux) | Zooms | Zooms | | Hold `Space` and drag | Pans | Pans | | Middle-click and drag | Pans | Pans | | Click and drag on empty canvas | Draws a selection box | Draws a selection box | Zoom runs from 1% up to 200%. Double-clicking the canvas never zooms, so a stray click won't move your view. Selection is covered in [Organizing the canvas](/docs/canvas/organizing). Scrolling over a chat you've clicked scrolls its conversation, not the canvas. A badge by the cursor shows which will happen: four arrows mean the canvas pans (Touchpad mode only), up and down arrows mean the chat scrolls. ## Viewport controls and mini map The controls in the bottom-right corner: - **Fit View** zooms and pans so every node fits on screen. It's also in the right-click menu under **View**. - **Zoom In** and **Zoom Out** step the zoom. The current zoom level sits between them. - **Show Mini Map** (the map icon) opens an overview of the whole canvas. Drag inside it to move your view, or scroll over it to zoom. Click the same button, now an X, to close it. ## Title pills when zoomed out Below 30% zoom, node titles are too small to read, so each node shows its title in a pill that stays the same size at any zoom. Groups keep their own labels. - **Click the title** to fly to that node. A node too tall for one screen lines up with the top of the screen, so you start at its beginning. - **Drag the grip** on the left of the pill to move the node from far out. It behaves like dragging the node itself, so the rest of your selection moves with it. {/* VIDEO: Zoom out on a busy canvas until title pills appear, click one to fly to its node, then drag another node by its grip. */} ## The node navigator The column of short lines on the left edge of the canvas is the node navigator, one line per node. The active node's line is longer and orange, and nodes that feed context into it are orange too. - **Hover** a line to see the node's name, type and connection counts, such as "2 in" and "1 out". - **Click** a line to make that node active and fly the canvas to it. - **Drag** a line to the right to add that node as context for the active node, or to remove it if it's already connected. See [Connections and context](/docs/canvas/connections). When the active node can take context, your groups are listed at the bottom. Clicking or dragging a group's line adds or removes the whole group as context instead of flying to it. To find a node by its content, use [Search](/docs/canvas/search) (`⌘K`, or `Ctrl+K` on Windows and Linux). ## Canvas header The bar across the top of the canvas, from left to right: - **Sidebar button** (**Hide Sidebar** / **Show Sidebar**): shows or hides your canvases and folders ([Canvases and folders](/docs/canvas/canvases-and-folders)). A dot on it means an app update is available. - **Breadcrumb**: the folder path and canvas name, such as "Research / Sleep science". An unnamed canvas shows "Untitled". - **Canvas settings** (gear icon): opens the settings below. - **Side view button** (panel icon): opens a panel on the right showing the active node at full height. See [Side view](/docs/canvas/side-view). ## Canvas settings Click the gear icon in the canvas header to open **Canvas settings**. Default model and Agent workspace belong to the canvas you're in. Everything else applies to every canvas. | Setting | What it does | Applies to | | --- | --- | --- | | **Movement** | **Touchpad (pan)**: two-finger scroll moves the canvas. **Mouse (scroll wheel)**: the wheel zooms. | All canvases | | **Pinch zoom speed** | Windows only. How fast pinch and `Ctrl` + scroll zoom, from 1 to 20 (default 8). | All canvases | | **Default model** | The model this canvas's chats use unless you pick another in the chat. **Global default** shows while it follows your app-wide default. See [Choosing a model](/docs/ai-features/models). | This canvas | | **Agent workspace** | The folder Claude Code, Codex and Cursor work in. **Choose folder...** sets it; the X clears it. A chat can also pick its own. See [Subscription providers](/docs/settings/subscription-providers). | This canvas | | **Auto-switch into Orchestrator mode** | Applies an agent's proposal to switch into Orchestrator mode without asking. Orchestrator mode sends canvas context to Slashspace's cloud. **Always allow** on a switch proposal turns this on. See [Orchestrator mode](/docs/ai-features/orchestrator). | All canvases | | **Edges** | Line style for connections: Default, Straight, Step, Smooth step or Simple bezier. See [Connections and context](/docs/canvas/connections). | All canvases | **Go to Settings** at the bottom of the popover opens the app's full settings. ## Next - [Adding nodes](/docs/canvas/creating-nodes) - [Organizing the canvas](/docs/canvas/organizing) - [Keyboard shortcuts](/docs/canvas/keyboard-shortcuts) # Keyboard shortcuts (https://www.slashspace.ai/docs/canvas/keyboard-shortcuts) Every keyboard shortcut in Slashspace for Mac, Windows and Linux, grouped by where it works. A long session goes faster when your hands stay on the keyboard: add a node, ask, branch and keep going without reaching for the mouse. On a Mac, `⌘` is Command, `⌥` is Option, `⌃` is Control and `⇧` is Shift. ## Canvas These work when you aren't typing in a text box. For trackpad and mouse gestures, see [Navigating the canvas](/docs/canvas). ### Move around | Action | Mac | Windows and Linux | |---|---|---| | Pan | Hold `Space` and drag | Hold `Space` and drag | | Zoom | `⌘` + scroll | `Ctrl` + scroll | ### Add nodes For the first seven rows, press the letter, then click where the node should go. `Esc` or the same letter again cancels. See [Adding nodes](/docs/canvas/creating-nodes). | Action | Mac | Windows and Linux | |---|---|---| | Chat node | `C` | `C` | | Text node | `T` | `T` | | Image node | `I` | `I` | | Scratch pad | `S` | `S` | | Post node | `V` | `V` | | Document node | `D` | `D` | | Web node | `W` | `W` | | Chat node in [Orchestrator mode](/docs/ai-features/orchestrator), placed mid-window and ready to type in | `A` | `A` | | Chat node right where you click | Hold `⌘` and click empty canvas | Hold `Ctrl` and click empty canvas | | Empty [branch](/docs/canvas/branching) of the active node | `⌘B` | Not available. Double-click the node's right handle instead. | ### Select and organize See [Organizing the canvas](/docs/canvas/organizing). | Action | Mac | Windows and Linux | |---|---|---| | Select every node | `⌘A` | `Ctrl+A` | | Add a node to the selection, or take it out | `Shift`-click | `Shift`-click | | Add a selection box to the current selection | `Shift`-drag on empty canvas | `Shift`-drag on empty canvas | | Clear the selection | `Esc` | `Esc` | | Tidy Up | `0` | `0` | | Delete the selected nodes and connections | `Delete` | `Backspace` or `Delete` | | Copy the selected nodes | `⌘C` | Not available. Use **Duplicate** in the node's header. | | Paste copied nodes, or turn clipboard text, links and images into nodes | `⌘V` | `Ctrl+V` | | Undo | `⌘Z` | `Ctrl+Z` | | Redo | `⌘⇧Z` | `Ctrl+Y` on Windows, `Ctrl+Shift+Z` on Linux | The `0` key arranges the canvas but leaves your view where it is; the toolbar's **Tidy Up** button also zooms to fit. Undo and redo include Scratch pad strokes. With the cursor in a text box, they undo and redo your typing instead. ## Chat composer These work in a chat node's message box and in the [side view](/docs/canvas/side-view), which has no shortcuts of its own. See [Chat node](/docs/nodes/chat-node). ### Send and branch | Action | Mac | Windows and Linux | |---|---|---| | Send | `Enter` | `Enter` | | New line | `Shift+Enter` | `Shift+Enter` | | Send into a new branch (**Submit in new branch**) | `⌘⇧Enter` | `Ctrl+Shift+Enter` | | Start an empty branch of this chat | `⌘B` | `Ctrl+B` | ### Modes Mode shortcuts act on the active chat (the one you last clicked), even when you aren't typing in it. See [Modes](/docs/nodes/chat-node#modes). | Action | Mac | Windows and Linux | |---|---|---| | Switch to **Agent** | `⌘1` | `Ctrl+1` | | Switch to **Chat** | `⌘2` | `Ctrl+2` | | Switch to **Image** | `⌘3` | `Ctrl+3` | | Switch to **Orchestrator** | `⌘4` | `Ctrl+4` | ### Dictation See [Voice input](/docs/ai-features/voice-input). | Action | Mac | Windows and Linux | |---|---|---| | Start dictating in an empty message box | `Space` twice, quickly | `Space` twice, quickly | | Start dictating with the cursor in the message box | `Ctrl+M` (Control, not Command) | `Ctrl+M` | | Finish and transcribe | `Enter` | `Enter` | | Cancel the recording | `Esc` | `Esc` | ### Slash and mention menus Type `/` to open the [slash menu](/docs/ai-features/slash-commands), or `@` to [mention a node](/docs/canvas/connections#mentions-create-connections). | Action | Mac | Windows and Linux | |---|---|---| | Move through the list | `↑` / `↓` | `↑` / `↓` | | Choose the highlighted item, or open a section | `Enter` or `Tab` | `Enter` or `Tab` | | Open the highlighted section (`/` menu only) | `→` | `→` | | Back out of a section | `←` or `Esc` | `←` or `Esc` | | Close the menu | `Esc` | `Esc` | ### Replies and edits | Action | Mac | Windows and Linux | |---|---|---| | Create the branch from the "Ask about this…" bubble on selected reply text (see [Ask about a passage](/docs/canvas/branching#ask-about-a-passage)) | `Enter` | `Enter` | | Close that bubble | `Esc` | `Esc` | | **Save & Rerun** in the **Edit Message** dialog | `⌘Enter` | `Ctrl+Enter` | ## Text formatting These work in a [Text node](/docs/nodes/text-node). Most also work in a chat's message box, where `⌘B` branches instead. | Action | Mac | Windows and Linux | |---|---|---| | Bold | `⌘B` | `Ctrl+B` | | Italic | `⌘I` | `Ctrl+I` | | Strikethrough | `⌘⇧S` | `Ctrl+Shift+S` | | Code | `⌘E` | `Ctrl+E` | | Heading 1, 2 or 3 | `⌘⌥1`, `⌘⌥2`, `⌘⌥3` | `Ctrl+Alt+1`, `Ctrl+Alt+2`, `Ctrl+Alt+3` | | Turn a heading back into normal text | `⌘⌥0` | `Ctrl+Alt+0` | | Bullet list | `⌘⇧8` | `Ctrl+Shift+8` | | Numbered list | `⌘⇧7` | `Ctrl+Shift+7` | | Blockquote | `⌘⇧B` | `Ctrl+Shift+B` | | Code block | `⌘⌥C` | `Ctrl+Alt+C` | ## Search dialog `⌘K` works anywhere on a canvas, even while you're typing. See [Search](/docs/canvas/search). | Action | Mac | Windows and Linux | |---|---|---| | Open or close search | `⌘K` | `Ctrl+K` | | Move through results | `↑` / `↓` | `↑` / `↓` | | Open the highlighted result | `Enter` | `Enter` | | Run the search (in **AI Search**) | `Enter` | `Enter` | | Close | `Esc` | `Esc` | ## App menu These come from the app menu and work anywhere in the app. For what to do with the logs, see [Troubleshooting and support](/docs/troubleshooting#logs-and-bug-reports). | Action | Mac | Windows and Linux | |---|---|---| | Open the log folder (**Help** > **Open Log Folder**) | `⌘⇧L` | `Ctrl+Shift+L` | | Cut, copy or paste text | `⌘X`, `⌘C`, `⌘V` | `Ctrl+X`, `Ctrl+C`, `Ctrl+V` | | Reload the window | `⌘R` | `Ctrl+R` | | Make the whole interface larger or smaller (not the canvas zoom) | `⌘+` or `⌘-` | `Ctrl++` or `Ctrl+-` | | Reset the interface size | `⌘0` | `Ctrl+0` | | Full screen on or off | `⌘⌃F` | `F11` | | Minimize the window | `⌘M` | `Ctrl+M` (in a chat's message box, it starts dictation instead) | | Quit | `⌘Q` | `Ctrl+W` closes the window, which quits | # Organizing the canvas (https://www.slashspace.ai/docs/canvas/organizing) Select, arrange, group, color and copy nodes so a sprawling canvas turns into a map you can come back to. Exploration gets messy by design. You open a source, ask a side question, fork an answer, and an hour later there are nodes everywhere. Tidying that sprawl into readable clusters is what turns a session into a map you can return to. Layout is for you, not the AI. Context follows connections, not position, so aligning or tidying never changes what a chat sees. ## Selecting nodes - Click a node to select it. `Shift`-click to add a node to the selection or take one out. - Drag across an empty part of the canvas to draw a selection box. Every node it touches is selected. Hold `Shift` while dragging to add to the current selection. - Press `⌘A` (`Ctrl+A` on Windows and Linux) to select everything. - Press `Esc`, or click an empty part of the canvas, to clear the selection. To move a node, drag it by its header, where a grip appears when you hover. Dragging any selected node moves the whole selection. When you're zoomed far out, drag a node's title pill instead (see [Navigating the canvas](/docs/canvas)). ## The selection toolbar Select two or more nodes and a toolbar appears above them: | Button | What it does | |--------|--------------| | **Align top** | Lines up the top edges with the highest node. | | **Distribute horizontally** | Spaces the left edges evenly between the leftmost and rightmost node. Needs three or more. | | **Distribute vertically** | Spaces the top edges evenly from top to bottom. Needs three or more. | | **Group selection** | Wraps the selection in a group. | The toolbar acts on loose nodes only. Groups, and nodes already inside one, are left out. {/* VIDEO: Drag a selection box around four scattered nodes, then click Align top and Distribute horizontally. */} ## Tidy Up **Tidy Up** rearranges the whole canvas into left-to-right columns that follow your connections. Nodes that nothing feeds into start the first column, and each connected node lands one column further right than the nodes feeding it. Sources end up left of the chats that use them, and branches fan out to the right. Unconnected nodes stack at the bottom of the first column, and each column keeps its current top-to-bottom order. A group moves as one block, with its contents untouched. To run it: - Press `0` (while you're not typing). - Click **Tidy Up** (the broom) in the toolbar at the bottom of the canvas. This also zooms to fit the result. - Right-click the canvas and choose **Tidy Up** under **View**. Prefer your old layout? One undo puts it back. A selected group has its own **Tidy up** button that arranges only the nodes inside it. {/* VIDEO: Press 0 on a messy canvas and watch it settle into left-to-right columns, then undo. */} ## Minimize and maximize Right-click the canvas and choose **Minimize All Nodes** to collapse every chat node into a card showing its last question and answer, so you see the shape of the canvas instead of walls of text. **Maximize All Nodes** expands them again. To collapse one chat, hover its header and click **Collapse**. Click it again, or **See More** on the card, to expand it. See [Chat node](/docs/nodes/chat-node) for the other header controls. ## Colors Chat nodes and Text nodes can be tinted, for example green for conclusions and orange for open questions. Hover the header, click **Change Color**, and pick a swatch under **Pick a color**. The first, neutral swatch removes the color. ## Duplicate, copy and paste **Duplicate:** hover a Chat, Text, Image or Scratch pad node's header and click **Duplicate** (**Duplicate Node** on chats). A copy with the same content appears to the right. Connections aren't copied. **Copy and paste:** select nodes and press `⌘C`, then point where you want them and press `⌘V` (`Ctrl+V` on Windows and Linux). Connections between the copied nodes come along, and you can paste into another canvas. If text is selected, `⌘C` copies the text instead. > **Note:** Copying nodes with the keyboard works on Mac only. On Windows and Linux, use **Duplicate**. ## Groups A group is a labeled container for a cluster. Select two or more nodes, click **Group selection**, and type a name over "Untitled". The group resizes to fit its contents, and dragging it moves everything inside. Select a group for **Ungroup** and **Tidy up**. See [Group](/docs/nodes/group-node) for more, including groups as context. ## Undo and redo | Action | Mac | Windows | Linux | |--------|-----|---------|-------| | Undo | `⌘Z` | `Ctrl+Z` | `Ctrl+Z` | | Redo | `⌘⇧Z` | `Ctrl+Y` | `Ctrl+Shift+Z` | Undo covers adding and deleting nodes and connections, moving (dragging, aligning, distributing, Tidy Up), grouping and ungrouping, reconnecting, Scratch pad strokes, and changes [Orchestrator mode](/docs/ai-features/orchestrator) makes. Colors, collapsing and resizing aren't undoable. With your cursor in a note or chat input, `⌘Z` undoes typing instead. The canvas keeps the last 100 steps. History resets when you switch canvases or restart the app. ### Deleting and restoring Select nodes and press `Delete` or `Backspace`, or click the trash button in a node's header. Their connections go too. A "Node Deleted" toast appears with a **Restore** button that brings back the nodes and their connections. After the toast is gone, undo does the same. ## Related - [Group](/docs/nodes/group-node) - [Connections and context](/docs/canvas/connections) - [Keyboard shortcuts](/docs/canvas/keyboard-shortcuts) # Search (https://www.slashspace.ai/docs/canvas/search) Jump to any node on the current canvas by typing what you remember, with Local Search or AI Search. On a large canvas you remember what was said, not where it sits. Search lets you type a few words you remember and fly straight to the node that holds them. Search covers the canvas you have open. To find a canvas by name, use the search box in the sidebar (see [Canvases and folders](/docs/canvas/canvases-and-folders)). ## Open search Press `⌘K` (`Ctrl+K` on Windows and Linux), or click the search icon at the right end of the bottom toolbar. The dialog always opens in **Local Search**; switch with the **AI Search** and **Local Search** buttons beside the search box. | Key | Action | | --- | --- | | `⌘K` (`Ctrl+K`) | Open or close search | | `↑` / `↓` | Move through results | | `Enter` | Open the highlighted result (in AI Search, run the search first) | | `Esc` | Close | Opening a result closes the dialog and moves the canvas to that node. If the match is in an AI reply, the chat also scrolls to that reply and highlights it briefly. {/* VIDEO: On a busy canvas, press ⌘K, type a half-remembered phrase, and press Enter to land on the matching chat reply. */} ## Local Search Local Search runs on your computer. With the search box empty, it lists every node on the canvas, most recently updated first, so it doubles as a node switcher. Type and the list narrows to matches, each with a match percentage. Matching is fuzzy, so typos and partial words still work. Titles count for more than content. | Node | Matched on | | --- | --- | | Chat node | Title, every message, and unsent text in the composer | | Text node | Title and note text | | Web node | Page title and URL | | Post node | Title and transcript | | Image node | File name | | Scratch pad | Title | Groups don't appear in results. To find text inside a document, use AI Search. If nothing matches, click **Try AI Search Instead**. ## AI Search AI Search matches by meaning, so you can describe an idea in words that never appear on the canvas, such as "the argument against my main claim". It runs on Slashspace's servers, and searching asks you to opt in to cloud indexing first. 1. Click **AI Search**. If you haven't opted in, you see "You have not opted in for cloud indexing!". Click **I understand, let me opt in** to continue. Switching to AI Search starts sending canvas text to Slashspace's servers before this check (see below). 2. Type what you're looking for and press `Enter`. 3. Each result shows the matching passage and a match percentage. Click one to jump to it. If nothing comes back, click **Try Local Search Instead**. ### What leaves your computer Each time you switch to AI Search, Slashspace sends the text of new or changed nodes on this canvas (chat messages, notes, captured web pages and post transcripts) to its servers, where it's kept in a search index for your account. Documents are already uploaded for processing when you add them, so their text is searchable too. The dialog's footer shows indexing progress and reads "done" when the index is current. Search before then and recent changes may be missing. For the full picture, including where to manage the opt-in, see [Privacy and your data](/docs/privacy). ## Related - [Navigating the canvas](/docs/canvas) - [Keyboard shortcuts](/docs/canvas/keyboard-shortcuts) # Side view (https://www.slashspace.ai/docs/canvas/side-view) Open one node in a full-height panel beside the canvas to read, write or keep chatting without losing your place. Long answers and long sources are hard to read inside a small node. The side view gives one node a full-height panel on the right of the window, while the canvas stays in view next to it. Keep it open to click through your sources one at a time, or to keep a conversation going while you pan the canvas for context to connect. {/* VIDEO: Click Open in sidebar on a chat node with a long answer, scroll it in the panel, then click a Document node and page through it. */} ## Open and close the side view - **From a node:** hover it and click **Open in sidebar** in its header. Every node type except groups has this button. (It opens the panel on the right, not the canvas list on the left.) - **From the canvas header:** click the panel button at the right end, next to the **Canvas settings** gear. Click it again to close the panel. - **From a Document node:** double-click the page preview. There is no keyboard shortcut for it. While the panel is open, it follows your selection: click another node and it switches. Click an empty part of the canvas, or select a group, and it shows "Select a node on the canvas to view its content here". ## What each node shows | Node | In the side view | |------|------------------| | Chat node | The whole conversation, with the node's full composer pinned to the bottom (mode, model, Tools menu, voice input, **Submit in new branch**). | | Text node | The note in a full-height editor with the formatting toolbar. | | Document node | Every page at the panel's width, with a page counter and up and down arrows. Files that aren't PDFs appear once processing has converted them. | | Image node | The image, with **Reveal in folder** and **Focus on Image canvas** (zooms the canvas to the node). | | Web node | The URL field with **Go** or **Refresh**, a screenshot when one was captured, and the **Parsed content** Slashspace extracted. | | Post node | The embedded post. Click **Ready** to read its transcript. | | Scratch pad | The pad and its toolbar, scaled to fit. You can keep drawing. | > **Note:** A Text node in the side view and the same node on the canvas don't refresh each other while you type. Edit in one place at a time so one copy doesn't overwrite the other. ## Resize the panel Drag the panel's left edge to resize it, anywhere from 280 to 640 pixels wide (it starts at 384). Slashspace remembers the width. The panel itself starts closed each time you open the app. ## Related - [Chat node](/docs/nodes/chat-node): the composer and message actions. - [Document node](/docs/nodes/document-node): processing, **Open Markdown** and **Open PDF**. - [Branching](/docs/canvas/branching): forking a thread from the composer. # Core concepts (https://www.slashspace.ai/docs/core-concepts) The mental model behind Slashspace: canvases, nodes, connections as context, branches, and modes. A chat app keeps everything in one scrolling thread, and the model reads whatever happens to be in it. Slashspace replaces the thread with a canvas. Your conversations, notes and sources sit side by side as nodes, and you draw connections to decide exactly what each conversation reads. Once the ideas below click, every other page is detail. ## Canvas A canvas is an infinite surface you pan and zoom around. Give each project or line of research its own canvas, so its sources and conversations stay together, and you come back weeks later to a map instead of a transcript. You can keep many canvases. They live in the sidebar, where you can sort them into folders (the sidebar calls canvases spaces). See [Canvases and folders](/docs/canvas/canvases-and-folders). ## Nodes A node is one unit of content and context: a conversation, a note, a web page, a video, a document, an image or a sketch. Anything you put on the canvas as a node is something a chat can read. | Node | What it holds | |------|---------------| | [Chat node](/docs/nodes/chat-node) | A conversation with an AI model. It reads whatever you connect to it. | | [Text node](/docs/nodes/text-node) | Your own notes, instructions or drafts, in rich text. | | [Web node](/docs/nodes/web-node) | The content of a web page, captured from its URL. | | [Post node](/docs/nodes/post-node) | A YouTube, TikTok, Instagram, Facebook, X or LinkedIn post, with its transcript when one is available. | | [Document node](/docs/nodes/document-node) | A PDF, Word, PowerPoint or Excel file, or a text or code file. | | [Image node](/docs/nodes/image-node) | An image from your computer, or one generated in Image mode. | | [Scratch pad](/docs/nodes/scratch-pad-node) | A freehand sketch. Connected chats see it as an image. | | [Group](/docs/nodes/group-node) | A labeled container for related nodes. Connect it to a chat to pass along everything inside. | ## Connections are context A connection decides what a chat reads. Every node has a handle on its right edge, and chat nodes also have one on their left. The right handle gives and the left handle takes: drag from a node's right handle to a chat's left handle, and the chat reads that node's content with every message you send. So context flows left to right: sources on the left, the conversations that use them to their right. Context also flows through chains. If A connects to B and B connects to C, the chat in C sees both A and B, not only its direct neighbor. A node that reaches a chat by more than one path counts once. You don't have to drag. Type `@` in a chat's message box and pick a node: the mention adds a connection from that node to the chat, and the connection stays after you send. See [Connections and context](/docs/canvas/connections). {/* VIDEO: Drag from a Web node's right handle to a chat's left handle, ask a question about the page, then connect that chat to a second chat and show it answering from both. */} ## Context modes Every chat node has a context mode, which sets what it sees beyond its own conversation. - **Summarized context** (the default). A chat with no connections gets a short summary of every other node on the canvas, so it knows what is there. In Agent mode, the AI can then open the full content of a node that looks relevant. Once you connect something, the chat reads only what is connected. - **Isolated context**. The chat ignores the rest of the canvas, including any connections, and works only from its own conversation. Use it when you want a clean-slate answer on a busy canvas. You switch modes from the chat's left handle while it has no connections. See [Connections and context](/docs/canvas/connections). ## Branches A long conversation drifts. You ask a side question, the answer takes over, and the thread you cared about gets buried. On a canvas you fork instead: click **Submit in new branch**, or press `⌘⇧Enter` (`Ctrl+Shift+Enter` on Windows and Linux), and your message starts a new chat to the right, connected to the one you were in. Because a branch is connected to the chat it came from, it sees that conversation and everything feeding it, while the original thread stays untouched. See [Branching](/docs/canvas/branching). ## Chat modes Each chat node runs in one of four modes, picked from the mode menu in its message box. The mode decides what the AI can do. | Mode | What it does | |------|--------------| | [Agent](/docs/ai-features/agent-mode) | The default. The AI can use tools: your connectors, sub-agents that work in their own chat nodes, and reading other nodes on the canvas. | | [Chat](/docs/nodes/chat-node) | A plain conversation with no tools. | | [Image](/docs/ai-features/image-generation) | Generates images from your prompt. Connected images can serve as references, and each result lands in its own Image node. | | [Orchestrator](/docs/ai-features/orchestrator) | Builds on the canvas for you. Give it a goal and it adds chat, text, web and post nodes, connects them, and edits their titles and text. | ## Where the AI runs Each chat node has its own model, chosen from the model menu in its message box, so two chats side by side can put the same question to different models. Out of the box, chats use **Auto**, which picks a Slashspace hosted model for each message. Slashspace's hosted models and hosted tools, such as image generation, are paid from your hosted wallet, and your plan includes some wallet balance. You can also use your Claude or ChatGPT subscription (through Claude Code or Codex), your own API keys from providers such as OpenAI, Anthropic and Google, local models through Ollama, or any OpenAI-compatible endpoint. None of these draw from the wallet. See [Choosing a model](/docs/ai-features/models). ## Where your data lives Slashspace is local-first. Each canvas, with its nodes, conversations and connections, is saved as a folder inside `SlashspaceOS` in your home directory, and your sidebar folders are real folders there. Content leaves your computer when a chat sends it to a cloud model, or when you use a hosted feature such as capturing a web page. See [Privacy and your data](/docs/privacy) for exactly what is sent and when. ## Next - [Adding nodes](/docs/canvas/creating-nodes): every way to put content on the canvas. - [Connections and context](/docs/canvas/connections): handles, chains, groups and context modes in detail. - [Chat node](/docs/nodes/chat-node): the message box, modes, models and tools. # Quickstart (https://www.slashspace.ai/docs) Install Slashspace, sign in, and spend ten minutes asking a question, branching the answer and giving a chat a source to read. Slashspace is an AI infinite canvas: a workbench for long, exploratory work like research, learning and writing. Instead of one scrolling chat thread, every conversation, note and source is a node you can place, connect and branch, so you can chase a tangent without losing the main line. [Why a canvas](/docs/why-a-canvas) makes the full case. ## Install Download Slashspace from [slashspace.ai/download](https://www.slashspace.ai/download): - **macOS:** Apple silicon or Intel - **Windows:** the `.exe` installer - **Linux:** the `.AppImage`, or the `.deb` for Debian and Ubuntu Run the installer, follow the prompts, and launch Slashspace. ## Sign in 1. Click **Sign in**. Your browser opens the Slashspace sign-in page, where you can sign in or create an account. 2. After you sign in, the browser sends you back to the app, which shows "Waiting for browser sign-in..." until then. If the app doesn't open, click **Open Slashspace** on the browser page. 3. Still stuck? Click **Copy login token** on the browser page, then **Paste login token** in the app. Paste it and click **Continue sign-in**. > **Note:** Coming from Rabbitholes? Sign in with the email you bought it with. If Rabbitholes is on this computer, the app first offers to import your canvases. See [Moving from Rabbitholes](/docs/migration). ## Choose a plan Slashspace has no free trial. There are two plans: the yearly subscription at $89 a year, or Lifetime at $249 once. Both are refundable within 7 days, no questions asked. If you already bought on the website, your plan is on your account: sign in with the email you bought with. If your account has no plan yet, a plan screen appears. Choose the yearly subscription or Lifetime there. Checkout opens in your browser, and the app continues once you finish. If it doesn't, click **Already subscribed? Refresh account**. Details are on [Plans and billing](/docs/plans-and-billing). ## Your first canvas On first sign-in you see **Welcome to Slashspace**. Click **Continue**, and Slashspace creates your first canvas (the sidebar on the left calls canvases spaces). The canvas holds one chat node, ready for your first message. It starts in **Agent** mode, where the AI can use tools, on the **Auto** model, which picks the right Slashspace model for each message. There's nothing to configure first. ## Try it: ask, branch, bring in a source {/* VIDEO: On a new canvas, ask a question, fork a follow-up with Submit in new branch, then paste a YouTube link, drag its right handle into a new chat node and ask about the video. */} ### 1. Ask a question Type a question in the chat node's message box and press `Enter` (`Shift+Enter` adds a new line). Pick something with room to explore, like "How did the printing press change the way scientists shared ideas?" Why it matters: the exchange lives in a node you can move and come back to, not in a transcript that scrolls away. ### 2. Fork a tangent Something in the answer catches your eye. Type a follow-up, but instead of sending it normally, click the branch icon beside **Send** (tooltip **Submit in new branch**) or press `⌘⇧Enter` (`Ctrl+Shift+Enter` on Windows and Linux). A new chat node appears to the right, connected to the first, and answers there. For an empty branch instead, press `⌘B` (`Ctrl+B` on Windows and Linux) while your cursor is in the message box. Why it matters: the branch builds on the conversation so far, while the original thread stays exactly as it was, so you can explore several directions side by side. See [Branching](/docs/canvas/branching). ### 3. Bring in a source 1. Copy the address of a web page or a YouTube video. 2. Click an empty spot on the canvas so no message box is active, then paste (`⌘V`, or `Ctrl+V` on Windows and Linux). - A web page address becomes a [Web node](/docs/nodes/web-node) with the address filled in. Click **Go** to capture the page. - A YouTube link becomes a [Post node](/docs/nodes/post-node), which fetches the video's transcript. 3. Drag from the handle on the new node's right edge to the handle on a chat node's left edge. Or drop it on an empty spot to get a new chat node that's already connected. 4. Ask that chat about the source, for example "What are the main claims here, and how well are they supported?" You can also type `@` in a message box and pick the node: mentioning a node connects it for you. Why it matters: connections decide what each conversation sees, so the answer rests on the source you chose, and the source sits next to the conversation that uses it. See [Connections and context](/docs/canvas/connections). ## Use your own AI (optional) **Auto** and the other models in the model picker's **Slashspace** group are hosted by Slashspace. They, and hosted tools such as image generation, are paid from your hosted wallet, and your plan includes some wallet balance. Your own AI never draws from the wallet: - **A Claude or ChatGPT subscription**, through Claude Code or Codex: [Subscription providers](/docs/settings/subscription-providers) - **Your own API keys** for OpenAI, Anthropic, Google and others: [API providers](/docs/settings/api-providers) - **Local models** on your computer: [Ollama](/docs/settings/ollama) Set them up in **Settings** (bottom of the sidebar) under **AI Settings**, then pick them from the model dropdown in any chat node. See [Choosing a model](/docs/ai-features/models). ## Where to go next - [Why a canvas](/docs/why-a-canvas): where chat threads break down. - [Core concepts](/docs/core-concepts): the mental model in one page. - [Connections and context](/docs/canvas/connections): control what each chat sees. - [Branching](/docs/canvas/branching): every way to fork a conversation. - [Chat node](/docs/nodes/chat-node): modes, models, tools and message actions. ## Get help Join the [Discord community](https://links.rabbitholes.ai/discord), email [support@slashspace.ai](mailto:support@slashspace.ai), or right-click the canvas and choose **Bug / Feedback** (also under **Settings** > **Troubleshoot**). For common problems, see [Troubleshooting and support](/docs/troubleshooting). # Moving from Rabbitholes (https://www.slashspace.ai/docs/migration) Bring your Rabbitholes canvases, settings and purchase into Slashspace, with your originals kept as a backup. Rabbitholes is now Slashspace, and your canvases and purchase come with you. Slashspace copies your work from the Rabbitholes folders on your computer (your originals stay put as a backup) and finds your purchase through the email you bought with. ## Before you start You need the email address you bought Rabbitholes with. You don't need your old license key or a receipt. If Slashspace finds a `rabbitholes` folder in your home folder that hasn't been imported yet, it shows the import screen first, even before sign-in. | What | In Rabbitholes | In Slashspace | | --- | --- | --- | | Canvases and files | `~/rabbitholes` | `~/SlashspaceOS` | | Settings | `~/rabbitholes/.config` | `Slashspace/config` in your app data folder | | Sign-in | `RabbitHolesAI` in your app data folder | `Slashspace` in your app data folder | `~` is your home folder (`C:\Users\you` on Windows). The app data folder is `~/Library/Application Support` on Mac, `%APPDATA%` on Windows and `~/.config` on Linux. ## Import your work 1. Download Slashspace from [slashspace.ai/download](https://www.slashspace.ai/download) and install it (it's a separate app from Rabbitholes). 2. Open Slashspace. The screen says "We found an existing RabbitHoles installation." Click **Migrate**. 3. The button changes to **Migrating...** while a checklist ticks off each step, from **Create directories** to **Finalize**. Then Slashspace restarts on its own. 4. Sign in (see below). 5. Open the **Everything** folder in the sidebar: that's where your Rabbitholes canvases are. You can rename it or move canvases out ([Canvases and folders](/docs/canvas/canvases-and-folders)). {/* VIDEO: Opening Slashspace on a computer with Rabbitholes, clicking Migrate, the checklist completing, and the app restarting with the Everything folder in the sidebar. */} There's no skip button. To start with an empty Slashspace instead, move or rename the `rabbitholes` folder before opening the app. ### If the import stops A failed step shows "Migration failed at" with the step's name and the reason, plus a **Retry** button. Retrying is safe: it skips anything already copied. If it fails again, email support@slashspace.ai with the step name and message. A damaged canvas file doesn't stop the import: Slashspace skips it and copies the rest. The list of skipped canvases may flash by before the restart, so if a canvas is missing, email support with its name. Its original is still in `rabbitholes`. ## Sign in with your purchase email The sign-in screen asks you to "Sign in with the email you used to purchase Slashspace AI (formerly RabbitHoles AI)." If you signed in to Rabbitholes on this computer, you may already be signed in. Otherwise, click **Sign in** and finish in your browser ([Quickstart](/docs#sign-in) has the details). **Asked to create an account?** That's expected if you only ever used a Rabbitholes license key. Create it with your purchase email, and your purchase links to it automatically. **Seeing the plan screen instead of your canvases?** Slashspace found no purchase on that email. Click **Sign out** and sign in with the email on your Rabbitholes purchase confirmation. Not sure which one? Email support@slashspace.ai. ## What carries over - **Canvases:** every canvas with its nodes, chats and connections. YouTube nodes become [Post nodes](/docs/nodes/post-node). - **Files:** images and documents are copied into `SlashspaceOS`, and your canvases point at the new copies. - **Settings:** default models, your model list, API keys, persona prompts, and MCP servers (listed in [Settings > Connectors](/docs/settings/connectors)). - **Sign-in**, if you were signed in to Rabbitholes on this computer. API keys move into your operating system's [secure storage](/docs/settings/api-providers#where-your-key-is-stored), and Slashspace then removes the plain-text copies from the old `rabbitholes` settings files. Apart from that, nothing in the Rabbitholes folders is changed or deleted. What doesn't carry over: your license key (signing in replaces it), Straico (no longer supported, so its key and any Straico default model are dropped), and RAG mode, which is retired. Chats read whatever you [connect to them](/docs/canvas/connections). ## Your lifetime deal Your lifetime deal stays a lifetime deal. **Settings > Plans** shows your plan's name (such as Lifetime) tagged **Lifetime updates**, with "Every future version is yours. Nothing to renew." You get the same app features as a current Slashspace license, with no canvas limit, and your own API keys keep working. A Rabbitholes subscription shows "Your plan continues unchanged." instead. If your plan is missing, click **Already subscribed? Refresh** at the bottom of the page. Slashspace also has optional hosted AI models, for when you'd rather not manage API keys. They're paid from the **Optional hosted wallet**, which also pays for hosted tools such as image generation. Your own keys never draw from it. Hosted usage is now shown in dollars. Rabbitholes counted it in per-message units; Slashspace charges what each message costs, so your balance won't match your old Rabbitholes number. For adding funds and redeeming codes, see [Plans and billing](/docs/plans-and-billing). ## Common questions **Can I delete the Rabbitholes folder?** Yes, once you've checked that a few canvases with images or documents load in Slashspace. **Do I need to keep Rabbitholes installed?** No. The apps use separate folders, so changes in one don't appear in the other. **Can I import again later?** No, the import runs once. Work you do in Rabbitholes afterwards stays there. # Chat node (https://www.slashspace.ai/docs/nodes/chat-node) Ask AI questions on the canvas, choose each chat's mode and model, and rerun, edit or fork any answer. A chat node is where you talk to AI on the canvas. Unlike a chat app, each chat sees only the context you connect to it, and any answer can be forked into a new branch, so a side question never buries the thread you care about. {/* VIDEO: Ask a question in a new chat node, hover the answer to show Copy and Regenerate, then open the mode menu and the + Tools menu. */} ## The message box A new chat node is an empty message box with the placeholder "Type / for commands or @ to mention nodes". Once the chat has messages, the box sits under the node and appears when you click it. - `Enter` sends. `Shift+Enter` adds a new line. - While the AI answers, **Send** turns into **Stop**. Sub-agents it starts have their own **Cancel** button (see [Sub-agents](/docs/ai-features/sub-agents)). - **Submit in new branch** sends your message into a new connected chat instead: click the split-arrows button next to **Send**, or press `⌘⇧Enter` (`Ctrl+Shift+Enter` on Windows and Linux). See [Branching](/docs/canvas/branching). - The microphone button dictates, and so does pressing `Space` twice in an empty box. See [Voice input](/docs/ai-features/voice-input). ## Slash commands and mentions - Type `/` to add a skill, a saved prompt or a persona, switch a connector on or off, or set the reasoning level. See [Slash commands, skills and prompts](/docs/ai-features/slash-commands). - Type `@` to mention another node or group on this canvas. The mention connects it to the chat as context, and the connection stays after you send. See [Connections and context](/docs/canvas/connections). ## Modes The mode button in the message box sets what the AI can do in this chat. New chats start in **Agent**. The mode is saved with the node and applies from your next message. Shortcuts act on the active chat (the one you last clicked). | Mode | Mac | Windows and Linux | What it does | |------|-----|-------------------|--------------| | **Agent** | `⌘1` | `Ctrl+1` | The default. The AI can work in several steps and use tools: your connectors, sub-agents, and reading other nodes on the canvas. See [Agent mode and tools](/docs/ai-features/agent-mode). | | **Chat** | `⌘2` | `Ctrl+2` | A plain conversation with no tools. | | **Image** | `⌘3` | `Ctrl+3` | Generates images from your prompt, each in its own connected Image node. See [Image generation](/docs/ai-features/image-generation). | | **Orchestrator** | `⌘4` | `Ctrl+4` | Give it a goal and it builds on the canvas for you, adding, connecting and editing nodes. See [Orchestrator mode](/docs/ai-features/orchestrator). | **Video** also appears in the menu, marked "Soon". ## Choosing a model The model button, next to the microphone, picks the model that answers in this chat, and your choice is saved with the node. New chats start on your default model (**Auto** unless you change it); see [Choosing a model](/docs/ai-features/models) for what's in the menu. In Image mode the button lists image models. Orchestrator mode hides it, because it runs on a model Slashspace manages. ## The Tools menu Click **+** (tooltip "Tools") at the left of the message box. These settings apply to this chat only. - **Connectors**: all your connected [connectors](/docs/settings/connectors) are on by default; uncheck one to turn it off here. **Manage connectors** opens Settings. Agent mode only; in other modes the entry reads "Available in Agent mode". - **Personalities**: apply one of your [personas](/docs/ai-features/personas), or choose **Add New** to write one. Click the active persona again to turn it off. Orchestrator mode ignores personas. - **Reasoning**: **Low**, **Medium** (the default), **High** or **Max**. Higher levels let models that support reasoning think longer before answering. ## Context A chat's left handle is where context comes in. Hover it to see how many connections feed the chat. With nothing connected, click it to switch between **Summarized context** (the chat gets a short summary of the canvas) and **Isolated context** (it works only from its own conversation). See [Connections and context](/docs/canvas/connections). ## Message actions Hover a message to show its actions. On your messages: - **Copy** copies the text. - **Rerun** sends the same message again. Everything after it is removed and the AI answers afresh. - **Edit** opens **Edit Message**. **Save** changes the text and keeps the rest of the conversation. **Save & Rerun** (`⌘Enter`, `Ctrl+Enter` on Windows and Linux) also removes everything after it and gets a new answer. On AI answers: - **Copy** copies the answer's text. - **Regenerate** gets a new answer in its place and removes any messages after it. > **Tip:** Rerun, Edit and Regenerate rewrite the thread. To try a different angle and keep what you have, send it with **Submit in new branch** instead. ## Tool calls and approvals In Agent mode, each tool the AI uses appears in its answer as a row you can open. When a tool needs permission, the row asks "Allow this tool to run?" with **Approve** and **Reject**, and the AI waits. Proposals to switch this chat's mode or model work the same way. See [Agent mode and tools](/docs/ai-features/agent-mode) and [Sub-agents](/docs/ai-features/sub-agents). ## The node header Hover a chat's header to show its controls. | Control | What it does | |---------|--------------| | **Change Color** | Tints the node. | | **Duplicate Node** | Adds a copy with the same conversation and settings to the right, without its connections. | | **Collapse** | Folds the chat into a card with your last question and answer. Click **See More** or **Collapse** again to expand it. | | **Open in sidebar** | Opens the chat in the [side view](/docs/canvas/side-view). | | **Delete Node** | Deletes the chat. The "Node Deleted" message offers **Restore**. | An untitled chat names itself after its first answer, using the **Title Generation Model** from Settings > **Default Models**. Click the title to rename it, or hover it and click **Generate Title** (the sparkles button) for a fresh one. ## When something goes wrong Errors appear as a banner at the bottom of the conversation. - **The hosted wallet is empty.** Slashspace's hosted models (including **Auto**) and hosted tools such as image generation draw from your hosted wallet. When it runs dry, you see "Your hosted wallet has insufficient funds for the Slashspace provider. Add funds or switch to BYOK, a subscription provider, or a local model. Your license is unchanged." Click **Add funds** to top up, or switch to a model that doesn't use the wallet (your own API key, Claude Code, Codex or Ollama). - **A plan is required.** If your plan has ended, or a feature isn't in your plan, the banner says so (for example "Your trial has ended. Your canvases are still here. Subscribe to keep using Slashspace.") and **Upgrade** shows your options. In Image mode without a plan that includes image generation, sending opens the "Upgrade to generate images" window. See [Plans and billing](/docs/plans-and-billing). - **Anything else.** A red banner shows the error with **Retry**, which asks for the answer again. When a Slashspace model is selected and less than $1.50 is left, the model button shows the balance (for example "$1.20 left") so you can act before a reply fails. For more fixes and how to reach support, see [Troubleshooting and support](/docs/troubleshooting). # Document node (https://www.slashspace.ai/docs/nodes/document-node) Put PDFs, Office files and text files on the canvas so connected chats can read them. A Document node turns a paper, report, slide deck or spreadsheet into a source the AI can read. It sits next to the conversations about it, so you can see which chats drew on which source. Add a long report once, connect it to three chats, and each can work through it from a different angle: one summarizes, one checks the numbers, one drafts a reply. ## Add a document - **Toolbar:** press `D` (or click **Document node**), click where the node should go, then pick one or more files. - **Right-click menu:** right-click the canvas and choose **Document Node** (marked **Beta**). - **Drag and drop:** drop files from your file manager onto the canvas. - **Paste:** copy a file, then paste it onto the canvas. Files Slashspace can't use show a message such as "File type .zip is not supported". {/* VIDEO: drop a PDF onto the canvas, wait for Ready, connect it to a chat and ask a question about it */} ## Supported files | Kind | Extensions | |------|------------| | PDF | `.pdf` | | Word | `.docx`, `.doc` | | PowerPoint | `.pptx`, `.ppt` | | Excel and CSV | `.xlsx`, `.xls`, `.csv` | | Text and Markdown | `.txt`, `.md`, `.rtf`, `.html`, `.json`, `.xml` | | Code and config files | `.py`, `.js`, `.ts`, `.sql`, `.yaml` and many more (drag and drop or paste only) | Keep documents under 50 MB. The file picker skips anything larger. ## How processing works When you add a document, Slashspace uploads a copy to its servers and converts it there into Markdown (plain text with headings and tables) and a simple PDF. This doesn't happen on your computer, so the file leaves your device. See [Privacy and your data](/docs/privacy). The node shows its progress along the bottom: | Status | Meaning | |--------|---------| | **Starting...** | The file is uploading. | | **Queued** or **Processing...** | The file is being converted. Larger files take longer. | | **Ready** | Done. The **.md** and **.pdf** buttons appear. | | A red status such as **Failed** or **Timed Out** | Processing stopped. See [If processing fails](#if-processing-fails). | PDFs show their first page right away, with a page count. Other files show a file-type card until processing finishes. Double-click the preview to read the whole document in the [side view](/docs/canvas/side-view). ## Open Markdown and Open PDF When a document is **Ready**, two buttons on the node open its converted copies in your web browser: - **.md** (**Open Markdown**): the converted text. This is what connected chats read, so check here that a table or slide came through. - **.pdf** (**Open PDF**): a plain PDF of that text. For Word, PowerPoint and Excel files, the node preview and side view show this version, so the original layout and images don't carry over. ## Use a document as context Connect the Document node's right handle to a chat node, or type `@` in a chat and pick the document. [Connections and context](/docs/canvas/connections) explains how context flows. A chat in **Isolated context** ignores its connections, documents included. Once the node shows **Ready**, a connected chat receives the full converted text with every message. Nothing is summarized or trimmed to snippets. Wait for **Ready** before you ask. Until then, Word, PowerPoint and Excel files are left out, and the chat tells you the file "wasn't included in context". ### Very large documents - If the converted text is larger than about 500 KB (very roughly 80,000 words), the chat leaves it out and tells the model it was omitted. Split long books or reports into smaller files, such as one per chapter. - The whole document is sent again with each message, so connect big documents only to the chats that need them. This leaves more of the model's context window for the conversation. ## If processing fails - **"Failed to start indexing this document."** The upload didn't go through. Check your connection and click **Retry**. - **A red status such as Failed or Timed Out.** Click the circular-arrow button next to it to process the file again. - **The node never reaches Ready**, or its status disappears (for example after you quit the app mid-processing). Delete the node and add the file again. - **A Word or PowerPoint file keeps failing.** Export it as a PDF and add that instead. Still stuck? See [Troubleshooting and support](/docs/troubleshooting). # Group (https://www.slashspace.ai/docs/nodes/group-node) Wrap related nodes in a labeled box you can move, tidy and hand to a chat as one piece of context. Exploratory work sprawls: the sources for one question in one corner, the chats arguing about them in another, a draft somewhere else. A group wraps one of those clusters in a labeled box, so you can move it, tidy it and give it to a chat as a single unit. Grouping alone doesn't change what any chat sees. A group becomes context only when you connect it. ## Create a group 1. Select two or more nodes (see [Selecting nodes](/docs/canvas/organizing#selecting-nodes)). 2. In the toolbar above the selection, click **Group selection**. 3. Click the name field above the group (it reads "Untitled") and type a name. This is the only way to make a group. Nodes already in a group don't count toward the selection, so groups can't be nested. {/* VIDEO: Select three nodes, click Group selection, name the group, then drag a fourth node into it and watch the group grow to fit. */} ## Work with a group - **Move:** drag the group by its dashed area. Everything inside moves with it. - **Add a node:** drag it over the group until the group highlights, then drop it. - **Remove a node:** drag it until it no longer overlaps the group, then drop it. - **Size:** the group hugs its nodes, growing and shrinking as they move or as a chat inside gets longer. There's no manual resizing. Branches you start from a node inside a group land in the same group. Click the group's dashed area to select it, and two buttons appear beside its name: - **Tidy up** arranges the nodes inside left to right, following their connections. The canvas-wide **Tidy Up** moves each group as one block and leaves its inside alone. - **Ungroup** removes the box and leaves the nodes where they are. Grouping, ungrouping and tidying each undo in one step. ### Deleting a group Selecting a group and pressing `Delete` or `Backspace` deletes the group **and every node inside it**. **Restore** on the "Node Deleted" toast, or undo, brings them back. To keep the nodes, click **Ungroup** instead. A group you've emptied stays behind until you select it and press `Delete`. ## Use a group as context Connect a group to a chat and the chat reads everything inside it: each node arrives as if you had connected it directly, plus a note naming the group and listing its nodes top to bottom, left to right. The group is read fresh on every send, so later additions count from the next message. To connect one: - Drag from the group's right handle to a chat node, or double-click the handle to start a new chat already connected to it. - Type `@` in a chat's input and pick the group. Groups are listed with their node count. Unnamed ones appear as "Untitled group". - With a chat active, drag the group right from the bottom of the node navigator on the canvas's left edge. Only the group's own nodes count. Anything feeding into them from outside the group doesn't come along, so connect it separately if the chat needs it. See [Connections and context](/docs/canvas/connections). {/* VIDEO: Type @ in a chat, pick a group, and show the new connection from the group to the chat before asking about its contents. */} ## Related - [Organizing the canvas](/docs/canvas/organizing) - [Branching](/docs/canvas/branching) # Image node (https://www.slashspace.ai/docs/nodes/image-node) Put screenshots, diagrams and photos on the canvas so chats can see them, and keep generated images next to the chats that made them. Images are context too. Put a screenshot of an error, a chart from a paper or a whiteboard photo on the canvas, connect it to a chat, and the model looks at the picture itself, next to your notes and sources. Image nodes also hold the results of [image generation](/docs/ai-features/image-generation). {/* VIDEO: paste a screenshot onto the canvas, connect it to a chat, and ask what it shows */} ## Add an image - **Toolbar:** press `I` (or click **Image node** in the toolbar), click the canvas, then click **Add From Computer**. Right-clicking the canvas and choosing **Image Node** works too. - **Drag and drop:** drop image files onto the canvas. Each file becomes its own node. - **Paste:** copy an image or a screenshot, then press `⌘V` (`Ctrl+V` on Windows and Linux) on the canvas. **Add From Computer** takes PNG, JPG, GIF and WebP files up to 50 MB; drag and drop also accepts SVG. The title starts as the file name; click it to rename. ## View and find the file Hover over the node for **Reveal in Folder**, **Duplicate**, **Delete** and **Open in sidebar** (the [side view](/docs/canvas/side-view)). To see an image larger, zoom the canvas or drag the node's left, right or bottom edge. | How you added it | Where the file is | |---|---| | **Add From Computer**, drag and drop | Where it already was. The node points at the original, so moving or deleting that file breaks the node. | | Pasted (screenshots, images copied from a browser) | `~/SlashspaceOS/images/pasted` | | Generated | An `images` folder inside the [canvas's folder](/docs/canvas/canvases-and-folders) | ## Use an image as context Connect an image you added to a chat like any other source: drag from the image's right handle to the chat, or type `@` in the chat's composer and pick the image by its title. The chat sends the image with each message. See [Connections and context](/docs/canvas/connections). The chat's model has to accept images. **Auto** picks an image-capable model when an image is connected, but some models are text-only (for example **GPT-OSS 120B**) and Slashspace doesn't warn you. If a model ignores the picture or returns an error, switch [models](/docs/ai-features/models). > **Note:** An image reaches the chat only if it is 20 MB or smaller and stored inside your home folder. Images on an external drive aren't sent. ## Generated images You don't generate inside an Image node. In a chat set to **Image** mode, each generation branches into its own connected Image node titled "Generated image", with the prompt underneath. [Image generation](/docs/ai-features/image-generation) covers prompts, models and cost. | What the node shows | What it means | |---|---| | "Waiting to start…" or "Generating image…" | On its way. Once it's generating, **Cancel** stops it. | | **Before I generate** | The model needs a decision. Click an option, or type in "Or tell me what to change…" and press `Enter`. | | The finished image | **Edit** opens a new Image-mode chat connected to it. The clipboard button copies it. "via your OpenRouter key" means your own key paid. | | **Upgrade to generate images** | Your plan doesn't include hosted image generation. See [Plans and billing](/docs/plans-and-billing). | | **Out of usage balance** | Your hosted wallet is empty. **Add funds** opens your account in the browser. | | **Image generation cancelled** or **Image generation failed** | You stopped it, or the line underneath says what went wrong. Adjust the prompt in the chat and send again. | To build on any image, generated or added, connect it into an Image-mode chat as a reference for the next generation. ## Start a new node from an image Drag from the right handle and let go on empty canvas. A **Create node** menu offers **Chat** (a new chat with the image already connected), **Image** (an empty Image node) and **Scratch pad** (a new [Scratch pad](/docs/nodes/scratch-pad-node)). Press `Esc` to close it. # Post node (https://www.slashspace.ai/docs/nodes/post-node) Put a YouTube, TikTok, Instagram, Facebook, X or LinkedIn video on your canvas and turn it into a transcript your chats can read and quote. Lectures, talks and interviews are some of the richest sources you can learn from, but a video is hard to search and harder to quote. A Post node puts the video on your canvas and fetches its transcript, so every chat connected to it works from what was actually said. ## Supported links | Platform | Links that work | |---|---| | YouTube | Videos, Shorts, live streams and `youtu.be` links | | TikTok | Video links and `vm.tiktok.com` short links | | Instagram | Posts, Reels and IGTV | | Facebook | Videos, Watch, Reels, stories and `fb.watch` links | | X | Post links on `x.com` or `twitter.com` | | LinkedIn | Posts, videos and `lnkd.in` links | Share links work as copied. Reddit isn't supported: a pasted Reddit link becomes a [Web node](/docs/nodes/web-node). ## Add a Post node The quickest way: copy a video link, click an empty spot on the canvas and paste (`⌘V`, or `Ctrl+V` on Windows and Linux). Slashspace recognizes the platform and creates a Post node under your pointer. To start from an empty node: 1. Press `V` or click **Post node** in the toolbar, then click the canvas. You can also right-click the canvas and choose **Post Node**. 2. Paste a link into the "Paste a video post URL" field. If you type it instead, press `Enter`. For other ways to add nodes, see [Adding nodes](/docs/canvas/creating-nodes). {/* VIDEO: Copy a YouTube lecture link, paste it onto the canvas, wait for Ready, then connect it to a chat and ask for a summary. */} ## What you get - **A player.** YouTube, Facebook, X and LinkedIn videos show a thumbnail with title and length; click to play inside the node. Instagram and TikTok use the platform's own embed. The link icon under the player opens the original post in your browser. - **A transcript.** The status under the player shows "Loading transcript..." and then **Ready**. Click **Ready** to read the full transcript and copy quotes. Videos with captions usually arrive within a minute. When there are no captions, Slashspace generates a transcript from the audio, which can take several minutes for a long video. Keep working in the meantime. The link is sent to Slashspace's servers, which fetch the captions or a generated transcript from third-party transcription services. See [Privacy and your data](/docs/privacy). Transcripts are included with every plan and don't draw from your hosted wallet. Without an active plan, the node shows **Failed** with "Active desktop access is required to fetch media transcripts." See [Plans and billing](/docs/plans-and-billing). ## Use it as context Connect the Post node to a chat node, or type `@` in a chat's composer and pick it. The chat receives the node's title and the full transcript. [Connections and context](/docs/canvas/connections) explains how context flows. For example, add the lecture videos from a course, connect them all to one chat and ask for a study guide of the key ideas, noting which lecture each comes from. Rename each node after its lecture first (click the title, which starts as "YouTube Post"), because the chat sees node titles, not video titles. Then [fork the chat](/docs/canvas/branching) to dig into one lecture without cluttering the guide. > **Tip:** Wait for **Ready** before you ask. Until then, a connected chat sees only the node's title. ## When a transcript fails The status changes to **Failed**. Click it to see the reason, then click **Retry**. Common causes: - **Nothing to transcribe:** the post has no video (a text or photo post), or no one speaks in it. - **The post isn't public:** private or deleted posts can't be fetched. - **A temporary service problem:** wait a minute and retry. # Scratch pad (https://www.slashspace.ai/docs/nodes/scratch-pad-node) Sketch by hand on the canvas and let connected chats see your drawing as an image. Some ideas are faster to sketch than to describe: a flowchart, a page layout, an equation. A scratch pad is a sheet of paper on your canvas. Connect it to a [Chat node](/docs/nodes/chat-node) and the AI sees your drawing as an image, next to everything else you've wired in. ## Add a scratch pad - Press `S`, or click **Scratch pad** in the toolbar at the bottom of the canvas, then click where you want it. The right-click menu doesn't list scratch pads. - Drag from the right handle of an image node or another scratch pad onto an empty spot, then pick **Scratch pad**. ## Draw Draw with a mouse, a trackpad, a touch screen or a stylus. Any press on the paper draws, so move the pad by its header. A stylus that reports pressure draws thicker when you press harder. With other pointers, line weight follows the speed of your stroke. The bar across the top of the pad holds, from left to right: | Control | What it does | |---|---| | **Undo** and **Redo** | Step through this pad's strokes, erases and clears. | | **Pen** | Draws. Pads start with the pen. | | **Eraser** | Removes any whole stroke it touches. | | **Black**, **Red**, **Blue**, **Green** | Ink color. | | **Thin**, **Medium**, **Thick** | Pen width. | | **Clear** | Empties the pad. | Strokes, erases and clears are part of the canvas undo history, so `⌘Z` (`Ctrl+Z` on Windows and Linux) undoes them too. The pad's **Undo** and **Redo** buttons light up only when the next step belongs to that pad, so they never reverse a node move by surprise. Scrolling and pinching over the pad still pan and zoom the canvas. {/* VIDEO: draw a quick flowchart on a scratch pad, connect it to a chat, and ask "Turn this into a numbered list of steps". */} ## Ask a chat about your sketch 1. Connect the pad's right handle to a chat node, or type `@` in the chat's composer and pick the pad. 2. Ask about it, for example "Turn this flowchart into steps". Each time the chat sends, the pad is turned into an image of your ink on its paper background and attached to the message. Nothing is stored, so if you change the drawing and ask again, the AI sees the new version. An empty pad sends nothing. Pads further up a [chain or inside a connected group](/docs/canvas/connections) count too. If the chat doesn't seem to see your sketch: - Make sure the pad is connected. An unconnected chat only knows the pad exists. - Pick a [model](/docs/ai-features/models) that accepts images. Slashspace doesn't check this, and a text-only model can't read the drawing. - Check that the chat isn't set to **Isolated context**, which ignores all connections. **Agent** and **Chat** mode see the drawing. **Orchestrator** mode doesn't, and **Image** mode can't use a pad as a reference image yet. ## Resize and save Drag the pad's right or bottom edge to add or remove paper. Your drawing never scales. Strokes past a shrunken edge are hidden and left out of the image, and they come back when you enlarge the pad. The canvas saves your strokes. Your tool, color and width aren't saved: a pad goes back to the black, medium pen when you reopen the canvas. To sketch while you read a chat, click **Open in sidebar** in the pad's header. The [side view](/docs/canvas/side-view) shows the same pad, scaled to fit. # Text node (https://www.slashspace.ai/docs/nodes/text-node) Write notes, briefs and drafts on the canvas, format them, and feed them to any chat as context. A text node holds your own words on the canvas: notes, instructions, a brief, pasted excerpts, a draft. Connect it to a chat and the chat reads it with every message, so you write something once and reuse it wherever it matters. ## Add a text node Press `T` (or click **Text node** in the toolbar), then click where you want it. You can also paste plain text onto the canvas: anything that isn't a link becomes a text node. See [Adding nodes](/docs/canvas/creating-nodes) for every other way. A new node is titled "Text Note" and shows "Start typing..." until you write. Click the title to rename it. The node grows taller as you write; drag its left or right edge to change the width. ## Write and format The toolbar at the top of the node has **Heading 1**, **Heading 2**, **Heading 3**, **Bold**, **Italic**, **Strikethrough**, **Code**, **Bullet list**, **Numbered list** and **Blockquote**. `⌘B` and `⌘I` (`Ctrl+B` and `Ctrl+I` on Windows and Linux) toggle bold and italic. Markdown shortcuts work as you type: | Type | You get | |---|---| | `#`, `##` or `###`, then a space | A heading | | `-` or `1.`, then a space | A bullet or numbered list | | `>`, then a space | A blockquote | | `**bold**`, `*italic*`, `~~strike~~`, `` `code` `` | Inline formatting | | Three backticks, then a space | A code block with syntax highlighting | | `---` | A divider | `Enter` starts a new line (in a chat node it sends). Pasted Markdown arrives formatted. Web addresses you type become links, and clicking one opens it in your browser. ## Use a note as context Drag from the text node's right handle to a chat node's left handle. The chat now gets the note's title and full text with each message, always the latest version. Drop the connection on empty canvas instead and you get a new chat already connected to the note. You can also type `@` in a chat's composer and pick the note by title, which adds the same connection. [Connections and context](/docs/canvas/connections) covers chains and context modes. {/* VIDEO: Write a short brief in a text node, drag its right handle to a chat's left handle, and ask a question the chat answers using the brief. */} Typical uses: - **Reusable instructions**, such as a style guide, wired into every chat that writes. - **A brief** with the goal, audience and constraints, shared by each chat on the project. - **Excerpts** pasted from sources, kept beside the chats that discuss them. - **A draft to critique**: ask what to cut, revise the note, ask again. > **Tip:** Give each note a clear title. It's how you find it in the `@` menu, and the chat sees it as the note's heading. Text nodes give context but don't take it: they have a right handle only. ## Ask about a passage Selecting text in a text node doesn't open the "Ask about this…" bubble; that works on a chat's answers (see [Branching](/docs/canvas/branching)). To ask about a passage, connect the note to a chat and quote the passage in your question. ## Other actions Hover the header for **Change Color**, **Duplicate**, **Open in sidebar** (the [side view](/docs/canvas/side-view), where you can read and edit the note) and **Delete**. Double-click the header to zoom to the node. # Web node (https://www.slashspace.ai/docs/nodes/web-node) Capture a web page on the canvas so connected chats read what the page actually says. A Web node holds a captured copy of a web page: a screenshot of the top of the page and its text. Connect it to a chat and the AI works from the real article instead of guessing from memory, and the source stays on the canvas next to every conversation that uses it. {/* VIDEO: paste an article URL into a Web node, wait for the screenshot, then connect it to a chat and ask about the article */} ## Add a Web node - Press `W` (or click **Web page node** in the toolbar), then click the canvas. - Right-click the canvas and choose **Web Node**. - Paste a web address on the canvas. Links to YouTube, TikTok, Instagram, Facebook, X and LinkedIn posts become a [Post node](/docs/nodes/post-node) instead. Pasting an address into the "Enter URL" field starts the capture right away. If you typed it, or the node came from a paste on the canvas, press `Enter` or click **Go**. The node shows "Scraping..." while it works. > **Tip:** Links containing characters like `#`, `@` or parentheses paste as a Text node. Paste them into an empty Web node's field instead. ## What gets captured The finished node shows the screenshot, and its title becomes the site's domain (click the title to rename it). Hover the screenshot and click **View content** to read the extracted text, with a link that opens the original page in your browser. If no screenshot came back, the text shows in the node itself. Slashspace's hosted service loads the page, not your browser, and returns the text and screenshot. You need to be signed in, and capturing doesn't draw from the [hosted wallet](/docs/plans-and-billing). See [Privacy and your data](/docs/privacy). ## Use it as context Drag from the Web node's right handle to a chat, or type `@` in a chat and pick the node. The chat receives the node's title and the extracted text, the same text you see under **View content**. It doesn't see the screenshot. See [Connections and context](/docs/canvas/connections). ## Refresh or change the page A capture is a snapshot and doesn't update when the page changes. To capture it again, click **Open in sidebar** in the node's header, then click **Refresh** in the [side view](/docs/canvas/side-view). You can also edit the address there to capture a different page, or click the link to open the original. ## When a capture fails A red message appears in the node and the address stays in the field, so you can fix it and click **Go** again. Common causes: - **Logins and paywalls.** The hosted service can't use your browser's sign-ins, so you get the login page or a preview. Copy the text into a [Text node](/docs/nodes/text-node), or save the page as a PDF and add it as a [Document node](/docs/nodes/document-node). - **Blocked or script-heavy sites** can come back empty or partial. Try an archived or print version of the page. - **Private addresses** on your company network or your own computer aren't reachable from Slashspace's servers. If the node returns to its empty state with no error, the page had nothing Slashspace could read. For other problems, see [Troubleshooting and support](/docs/troubleshooting). # Plans and billing (https://www.slashspace.ai/docs/plans-and-billing) Pick a plan, see what the optional hosted wallet pays for, and manage updates, devices and billing. Your plan pays for the Slashspace app: using it on your computers and getting new versions. AI usage is separate. If you work with your own API keys, a Claude or ChatGPT subscription, or local models, the plan is all you pay Slashspace. The hosted wallet is optional: it pays for Slashspace's own hosted models and AI tools, which need no provider accounts. Use a model you already pay for, the wallet, or both, whichever costs you least. ## The plans | | Yearly subscription | Lifetime | |---|---|---| | Price | $89 a year, renews yearly | $249, one payment | | Updates | Every update while you subscribe | Every update we ship | | Devices | 2 | 2 | | Included hosted usage | $5, added once | $5, added once | Both plans include every feature and unlimited canvases. The included $5 goes into your hosted wallet when you buy, and isn't added again at renewal. ### Buying and the 7-day refund Slashspace has no free trial. Both plans are paid at checkout and are refundable within 7 days, no questions asked (see [Refunds](#refunds)). You don't need an account to buy. On [slashspace.ai](https://www.slashspace.ai/pricing), enter your email, answer three quick questions and pay on Stripe. Then download the desktop app and sign in with that same email. Your plan is already on the account. The yearly subscription renews every year until you cancel. Cancel any time from **Manage subscription** or your account page, and access continues to the end of the period you paid for. > **Tip:** If you have a yearly subscription and then buy Lifetime, the subscription ends at the close of the period you paid for, so it doesn't renew. ## Updates and when a plan ends - **Yearly subscription:** every new version installs while it's active. If you cancel, access continues to the end of the period you paid for and nothing more is charged. If a renewal payment fails, you have 7 days to update your card. - **When a subscription ends,** Slashspace opens to a plan screen instead of your canvases. Your canvases stay on your computer. Subscribe or buy Lifetime to open them in Slashspace again. - **Lifetime:** every future version is yours, with nothing to renew. ### If you have the earlier one-off license If you bought the earlier $89 one-off license, the app is yours for good and **Settings > Plans** shows **Updates until** a date. After that date your version keeps working, but new versions aren't offered. To get them, click **Subscribe for $89/year**, or **Get lifetime updates** to switch to Lifetime. ## The hosted wallet The **Optional hosted wallet** pays for: - **Slashspace's hosted models:** the **Slashspace** group in the [model picker](/docs/ai-features/models), including **Auto**. - **Hosted AI tools** such as [image generation](/docs/ai-features/image-generation), unless you use your own OpenRouter key for images. It never pays for models on your own [API keys](/docs/settings/api-providers), [subscription providers](/docs/settings/subscription-providers) (Claude Code, Codex, Cursor), [Ollama](/docs/settings/ollama) or [custom providers](/docs/settings/custom-providers). Capturing [web pages](/docs/nodes/web-node), fetching [post transcripts](/docs/nodes/post-node) and using included connectors don't draw from it either. Each hosted reply or image is charged what it cost to run, rounded up to the cent. A hosted request needs at least $0.50 in the wallet to start. Your plan adds 50 cents on top of the included $5 to cover that, so a new wallet shows a little over $5. ### Add funds 1. In **Settings > Plans** or **Settings > Account**, click **Add funds** on the wallet card. Your account page opens in your browser. 2. Click **Add funds** there and enter the **Spendable balance to add**: steps of $10, up to $200 per purchase. 3. Check **You pay** and click **Pay**. Each $10 of balance costs $11: the 10% platform fee is added on top, so the whole $10 is spendable. Adding funds opens once you've paid for Slashspace. Added funds don't expire and aren't refundable. ### When the wallet runs low or empty With a hosted model selected and less than $1.50 left, the model button shows the balance (for example "$1.20 left"). When the wallet can't cover a request, hosted replies stop with "Your hosted wallet has insufficient funds for the Slashspace provider." and image generation shows **Out of usage balance**. Nothing else changes: the app, your canvases and every non-hosted model keep working. Add funds, or switch the chat to another model. ## Manage your plan in Settings **Settings > Plans** shows your plan and its status, the plan cards if you don't have one yet, and the wallet. **Settings > Account** shows the same plan with a status chip, plus the wallet and your devices ([Account and devices](/docs/settings/account-management) explains the chips). Purchases and billing open in your browser. On a subscription, **Manage subscription** opens Stripe's billing portal to update your card, cancel or download invoices. Other plans get **Manage Billing**, which opens your account page on slashspace.ai. If a purchase you made in the browser doesn't show up, click **Already subscribed? Refresh** at the bottom of **Settings > Plans**. ## Devices Each plan covers 2 computers. Once you've paid for Slashspace, you can buy extra device slots for $49 each, a one-time purchase, with **Add slots** on your [account page](https://www.slashspace.ai/account). To see, rename or remove devices, go to [Account and devices](/docs/settings/account-management). ## Redeem a code There's no box to type a code into. Lifetime deals bought through a partner, and Rabbitholes lifetime deals, are matched to your account by the email you bought with. Sign in at [slashspace.ai/account](https://www.slashspace.ai/account) with that email (create the account if you need to) to see your plan, then sign in to the desktop app with the same email. If your deal doesn't show up, email **support@slashspace.ai** with your code and the email you bought with, and we'll apply it. ## Refunds You can ask for a refund within 7 days of a Lifetime purchase or of any yearly payment, no questions asked. Email **support@slashspace.ai** with your purchase details. Transaction fees are deducted, and a refund ends the access it paid for. Wallet top-ups aren't refundable. The [Terms of Use](https://www.slashspace.ai/legal/terms) have the full policy. ## Older plans If you're on a plan Slashspace no longer sells, such as an earlier monthly or yearly plan, it continues unchanged at its original price. **Settings > Plans** shows its name with **Current plan** and **Manage Billing**. Paid older plans can add wallet funds and device slots too. Rabbitholes lifetime deals keep lifetime updates: see [Moving from Rabbitholes](/docs/migration). ## Related - [Account and devices](/docs/settings/account-management) - [Choosing a model](/docs/ai-features/models) - [Image generation](/docs/ai-features/image-generation) # Privacy and your data (https://www.slashspace.ai/docs/privacy) What stays on your computer, what Slashspace sends elsewhere and when, and how to keep sensitive work local. Slashspace is local-first: your canvases are ordinary files on your computer, and Slashspace doesn't sync or back them up to its servers. Many features still need a server, though, and some send more than you might expect. Long-context work often involves unpublished drafts and sensitive sources, so it helps to know exactly where each piece goes. The [privacy policy](/legal/privacy-policy) is the formal version and names every service provider involved. ## What stays on your computer - **Canvases.** Each canvas is a folder inside `SlashspaceOS` in your home folder, holding its chats, notes, connections, captured text and generated images. See [where canvases live on disk](/docs/canvas/canvases-and-folders#where-canvases-live-on-disk). - **Prompts and skills.** Your prompt library is in `~/SlashspaceOS/.prompts` and installed skills are in `~/.agents/skills`. Personas and settings are in the app's data folder. - **API keys.** Keys are encrypted with your operating system's secure storage (the Keychain on a Mac) and don't sync. Each key goes only to its own provider, except with the OpenRouter image option below. Custom connector sign-ins are encrypted too, but environment variables for connectors that run on this computer are saved unencrypted. - **Local model runs.** Chats with an [Ollama](/docs/settings/ollama) model run on your computer. Chat titles, analytics and sign-in checks still use the network. - **Update backups.** When the app downloads an update, it copies your `SlashspaceOS` folder into a `backups` folder in the app's data folder, keeping the last few copies. ## What leaves your computer "Slashspace's servers" means Slashspace's own backend, which passes requests on to the providers named in the privacy policy. | Feature | What is sent | Where it goes | When | | --- | --- | --- | --- | | Chats on a Slashspace model, including **Auto** | Your message, earlier turns, the chat's context (connected nodes, or a summary of the canvas when nothing is connected), attachments and tool results | Slashspace's servers, then through Vercel AI Gateway to the model's maker, such as Anthropic or OpenAI. **Auto** also sends the start of your message to a small model that picks one. | Every send | | Chats on your own API key or a custom provider | The same | Straight to that provider or endpoint. With Qolaba, attached files are first uploaded to Slashspace's storage and Qolaba gets a link. | Every send | | Claude Code, Codex, Cursor | Your messages, canvas context and the tools Slashspace gives them | Anthropic, OpenAI or Cursor, under your own account | Every send | | Chat titles | The first message of each new chat | By default, Slashspace's servers and an OpenAI model, whatever model the chat uses | Each new chat | | Image descriptions | An image you add with **Add From Computer** | Your default chat model, when it runs on one of your API keys | When you add it | | [Orchestrator mode](/docs/ai-features/orchestrator) | Your request, a summary of the whole canvas (including where your documents are on your computer), nodes it opens and the earlier conversation | Slashspace's cloud, which runs it on an Anthropic model | Every message, whatever is connected | | [Image generation](/docs/ai-features/image-generation) | Prompt, settings and reference images | Slashspace's servers and storage, then OpenRouter. The result is stored there and copied to your computer. **Use my OpenRouter key for image generation** also sends your OpenRouter key. | Each generation | | [Web node](/docs/nodes/web-node) | The page address | Slashspace's servers, which fetch the page through a scraping service. That service hosts the screenshot. | When you capture a page | | [Post node](/docs/nodes/post-node) | The post link | Slashspace's servers and third-party transcription services | When you add the link | | [Document node](/docs/nodes/document-node) | The whole file | Slashspace's storage, a third-party parser, and a vector database of search embeddings. **Open Markdown** and **Open PDF** are public links: anyone who has one can open the converted file. | When you add it | | [AI Search](/docs/canvas/search) | Text of new or changed nodes on the canvas, and your searches | Slashspace's servers, kept in a search index for your account | Each time you switch to AI Search | | [Voice input](/docs/ai-features/voice-input) | The recording | Slashspace's servers, then Groq. With your own Groq key, straight to Groq. | When you finish recording | | [Connectors](/docs/settings/connectors) | Tool requests and results | Included connectors: Composio, set up through Slashspace's servers. Custom connectors: the server you added. | When a tool runs | | [Agent skills](/docs/settings/skills) | Browse and install requests | skills.sh | When you browse or install | | Sign-in and devices | Your email, a device ID, your computer's name and operating system, the app version | Slashspace's servers | At sign-in and in account checks | | Update checks | Your IP address, app version and platform | slashspace.download | At launch and when you check | | **Bug / Feedback** | Your message, attachments and the app's log, with file paths and email addresses removed | Slashspace's support team | When you send | Two things are easy to miss. Choosing **Orchestrator** mode is your go-ahead to send canvas context to Slashspace's cloud, and no notice appears when you switch to it yourself. And Claude Code and Codex can read any node on the canvas with a tool, even when the chat is set to **Isolated context**. ### Analytics and error reports Usage analytics go to PostHog, and they are not all off by default: - **AI usage**, for every chat on every model, including your own keys and Ollama: the model, provider, token counts, timings, tool names and any error. Never the text of your prompts or the replies. Always sent. - **Feature usage**: app launches, screens, clicks and feature events. A click can record the text you clicked, such as a canvas name. Sent before you sign in; after that, only while the analytics opt-in is on. **Error reports** go to Sentry and can include your IP address and any file path or canvas name in the error. About 1 in 10 sessions, and any session with an error, also send a screen replay with text, form fields and images masked. Once you sign in, analytics are linked to your account. No setting turns off AI usage analytics or error reports. ## Privacy settings **Privacy Settings**, at the bottom of **Settings > Account**, lists opt-ins, each with an **Enable** or **Disable** button. The list comes from Slashspace's servers, so it appears only while you're signed in and its wording can change. It currently has two: - Cloud indexing, the cloud vector database behind AI Search. With it off, AI Search shows "You have not opted in for cloud indexing!" instead of results. Switching to AI Search starts sending canvas text for indexing either way, and turning the opt-in off doesn't delete text already indexed. - Analytics, which covers the feature-usage analytics above. Both are off for a new account. Accounts that date from Rabbitholes keep the choices made there. Neither setting affects the rest of the table, AI usage analytics or error reports. ## Keeping sensitive work local - Chat on [Ollama](/docs/settings/ollama) or [your own API keys](/docs/settings/api-providers), not Slashspace models. - In **Settings > Default Models**, set **Title Generation Model** to a model on your own API key, so new chats aren't titled through Slashspace. - Keep sensitive material on its own canvas, and don't use AI Search or Orchestrator mode there. - Paste the passages you need into a [Text node](/docs/nodes/text-node) instead of adding a Document, Web or Post node. - Connect only what a chat needs. A chat with nothing connected gets a summary of every node on the canvas. **Isolated context** turns that off, though Claude Code and Codex can still read nodes. See [context modes](/docs/canvas/connections#context-modes). - If you dictate, save a Groq key so recordings skip Slashspace's servers. - Don't share **Open Markdown** or **Open PDF** links. ## Deleting your data **On your computer.** Delete a canvas from the sidebar (this is permanent), or delete the whole `SlashspaceOS` folder. Update backups stay in `~/Library/Application Support/Slashspace/backups` on a Mac and `%APPDATA%\Slashspace\backups` on Windows. If you moved from Rabbitholes, the old `~/rabbitholes` folder is still there too. **On Slashspace's servers.** Deleting a canvas doesn't remove copies already uploaded: documents, the AI Search index, and generated and reference images. To have your hosted data and your account deleted, email **support@slashspace.ai** from the address you sign in with. ## Related - [Privacy policy](/legal/privacy-policy) - [Canvases and folders](/docs/canvas/canvases-and-folders) - [Choosing a model](/docs/ai-features/models) - [Account and devices](/docs/settings/account-management) # Account and devices (https://www.slashspace.ai/docs/settings/account-management) Check which account and plan this computer uses, manage the computers that share your plan, and sign out. Your canvases live on your computer, but your plan and device slots belong to your account. **Settings > Account** shows both, so it's the first place to look when the app doesn't match what you bought. ## Open Account settings Click **Settings** at the bottom of the sidebar, then **Account** under **General**. The top card shows the email you're signed in with, a shortened **Device ID** for this computer, and **Sign out**. **Refresh account** sits above it on the right. ## Current plan The **Current Plan** card shows your plan's name, a status chip such as **Subscribed**, **Cancelled** or **Lifetime updates**, and the relevant date (renewal, or when a cancelled plan's access ends). The button in the card's corner depends on how you pay: - **Manage subscription** appears if you pay by subscription. It opens Stripe's billing portal in your browser, where you can update your card, cancel, or download invoices. - **Manage Billing** appears for every other plan (Lifetime, an earlier one-time purchase, or an older plan). It opens your account page on slashspace.ai. Without an active plan, the card says why and offers **Subscribe**. Below it, the **Optional hosted wallet** card shows your hosted balance in dollars. Prices, refunds and the wallet are covered in [Plans and billing](/docs/plans-and-billing). ## Devices Each computer you sign in on uses one of your plan's device slots. Slashspace recognizes the computer itself, so signing out and back in, or reinstalling the app, on the same computer doesn't take another slot. The **Devices** card lists each computer by name (or a shortened ID) with the date it was added. The header shows how many slots you're using, such as "1 of 2 devices". ### Manage devices on the web Removing devices happens on the web. Click **Manage devices on web** at the bottom of the Devices card (or go to [slashspace.ai/account](https://www.slashspace.ai/account) if the card isn't showing) and sign in with the same email if asked. In the **Devices** section there: - Click a device to open **Manage device**. Rename it under **Name** and click **Save**, or click **Remove device** to free its slot. - Click **Add slots** to add device slots. It appears once you've paid for Slashspace. ### When all your device slots are in use If every slot is taken when you sign in on a new computer, the sign-in screen shows "All your device slots are in use". To swap an old computer for the new one: 1. Click **Remove a device in account settings, then sign in again.** Your account page opens in the browser. 2. Click the old computer under **Devices**, then **Remove device**. 3. Back in Slashspace, click **Sign in** again. If you've already paid, you can click **Already own Slashspace? Buy an extra device slot** instead. {/* VIDEO: Hitting the device limit at sign-in, removing an old computer on slashspace.ai/account, then signing in again. */} ## Refresh your account Slashspace re-checks your plan on its own when you come back from checkout or billing in your browser. If the **Current Plan** card still looks out of date after a purchase, a plan change or a deal applied to your account, click **Refresh account**. If it can't refresh, check your connection, or sign out and sign back in. ## Sign out **Sign out** signs this computer out and returns you to the sign-in screen. Your canvases stay on your computer. Signing out doesn't free the computer's device slot: remove it on the web for that. ## Privacy settings **Privacy Settings**, at the bottom of the page, lists optional features that use your data, each with its own **Enable** or **Disable** button. [Privacy and your data](/docs/privacy) explains what each one controls and what leaves your computer either way. ## Delete your account To delete your Slashspace account, email **support@slashspace.ai** from the address you sign in with. Your canvases are files on your computer, so they stay where they are ([Canvases and folders](/docs/canvas/canvases-and-folders) shows where). ## Related - [Plans and billing](/docs/plans-and-billing) - [Troubleshooting and support](/docs/troubleshooting) # Add to Slashspace button (https://www.slashspace.ai/docs/settings/add-to-slashspace) Give Slashspace users a one-click link that opens your remote MCP server in the app, prefilled and ready to review. If you publish a remote MCP server, an Add to Slashspace link saves your users from copying URLs into settings. One click opens Slashspace on **Settings > Connectors** with your server's name and URL filled into **Add a custom connector**. Nothing is added, and Slashspace doesn't contact your server, until the user reviews it and clicks **Add**. ## Link formats | Link | Use it when | | --- | --- | | `https://www.slashspace.ai/mcp/add?name=...&url=...` | Websites and READMEs. The landing page works whether or not the visitor has Slashspace. | | `slashspace://mcp/add?name=...&url=...` | You know the user runs Slashspace 3.1.1 or later. Opens the app directly. | Both take the same two parameters: | Parameter | Value | | --- | --- | | `name` | Display name for the connector, up to 100 characters. | | `url` | Your MCP endpoint as an absolute `https://` URL, up to 2,048 characters. No username, password or `#fragment`. | Each must appear exactly once; any other parameter makes the link invalid. Build links with `URLSearchParams` so both values are percent-encoded: ```js const params = new URLSearchParams({ name: "Acme", url: "https://mcp.acme.com/mcp", }); const webLink = `https://www.slashspace.ai/mcp/add?${params}`; const appLink = `slashspace://mcp/add?${params}`; ``` ## Add the button HTML (note the `&` between parameters): ```html Add to Slashspace ``` Markdown: ```md [Add to Slashspace](https://www.slashspace.ai/mcp/add?name=Acme&url=https%3A%2F%2Fmcp.acme.com%2Fmcp) ``` Many Markdown renderers strip custom schemes like `slashspace://`, so use the `https://` form there. There is no official badge image yet; style the link like your other buttons. ## What your users see 1. **The landing page** shows "Add Acme to Slashspace", your server address and an **Open Slashspace** button (the browser may ask permission to open the app). Visitors without the app click **Download Slashspace**, install it, then return to the tab and click **Open Slashspace**. 2. **Slashspace opens.** If the user isn't signed in or doesn't have an active plan yet, they finish that first. The request waits as long as the app stays open. 3. **The dialog appears** on **Settings > Connectors** with the name and URL filled in. The user can edit either field or add headers. 4. **The user clicks Add.** Slashspace connects to your server and lists its tools, then shows "Connector added and checked." If your server needs sign-in, the dialog shows **Connect your account** with a **Sign in** button. If the connection fails, it shows "Could not connect" and offers **Save anyway**. **Cancel** discards the request. A link never overwrites an existing connector: if the name is taken, the dialog says "A connector with this name already exists" and the user renames it. Repeat clicks on the same link give one request; links for different servers queue (up to 10). ## Remote servers only A link carries only a name and an HTTPS URL. Slashspace tries Streamable HTTP first, then SSE. Local (stdio) servers can't be installed by link: publish the command, arguments and environment variables so users can enter them under **Runs on this computer** (see [Custom MCP connectors](/docs/settings/mcp-servers)). ## Authentication and secrets Never put API keys, tokens or other secrets in the link, including your server URL's query string. Links end up in browser history, logs and screenshots, and every user would share one credential. Instead, use: - **OAuth.** Answer unauthenticated requests with `401` and a `WWW-Authenticate` header, or publish `/.well-known/oauth-protected-resource` at your server's origin. Slashspace then offers **Sign in** and finishes in the user's browser. It uses dynamic client registration (a public client named "Slashspace") with PKCE and redirects to `http://127.0.0.1:/connectors/oauth/callback`. The port can change, so accept loopback redirects on any port. - **Headers.** Tell users which header to set (for example `Authorization`). They add it with **Add header** before clicking **Add**. ## Test your link 1. Open the `https://` link in a browser. "This connector link is incomplete or invalid" means a parameter is missing, repeated, not HTTPS or too long. 2. Click **Open Slashspace** and check the dialog shows your name and URL exactly. 3. Click **Add**, confirm the connector appears under **Your connections**, then remove it from its **⋯** menu. To try the app link without the landing page: ```sh # macOS open "slashspace://mcp/add?name=Acme&url=https%3A%2F%2Fmcp.acme.com%2Fmcp" # Linux xdg-open "slashspace://mcp/add?name=Acme&url=https%3A%2F%2Fmcp.acme.com%2Fmcp" # Windows (PowerShell) Start-Process "slashspace://mcp/add?name=Acme&url=https%3A%2F%2Fmcp.acme.com%2Fmcp" ``` An invalid app link shows a "Cannot add connector" error. If Slashspace opens with no dialog, it's older than 3.1.1 and needs an update. ## Related - [Custom MCP connectors](/docs/settings/mcp-servers): adding a server by hand. - [Connectors](/docs/settings/connectors): managing connectors. - [Agent mode and tools](/docs/ai-features/agent-mode): how chats use connector tools. # API providers (https://www.slashspace.ai/docs/settings/api-providers) Add your own API keys for OpenAI, Anthropic, Google and other providers, and pay each provider directly for what you use. Bring your own API keys to use any provider's models on the canvas. The provider bills you at its own prices, nothing comes out of the hosted wallet, and your conversations go from your computer straight to the provider. It's how you run a model Slashspace doesn't host, or put a long research session on an account you already pay for. Open **Settings** and choose **API Providers** under **AI Settings** (**Manage AI Models** at the bottom of the model picker opens it too). It lists OpenAI, Anthropic, Google, Perplexity, xAI, OpenRouter, Qolaba, DeepSeek, Groq and Fal AI (BETA). A green shield on a row means a key is saved on this computer. Slashspace's own hosted models need no key; see [Choosing a model](/docs/ai-features/models). ## Add a key 1. Click a provider to expand it. 2. Click **Get API key** on its row. The provider's key page opens in your browser: create a key and copy it. 3. Paste it into the "Add API key" field. 4. Click the check mark to save (the red X cancels). Slashspace checks the key with a small request to the provider before saving. The panel then shows **Securely stored on device**, and most providers get a starter set of models, listed in the model picker under the provider's name. OpenRouter starts with none, so add the ones you want. {/* VIDEO: Expand OpenAI in Settings > API Providers, click Get API key, paste a key and save, then pick one of its models in a chat node. */} ### If the key is rejected A failed check shows "Could not validate the API key. Check the key and your connection, then retry." (OpenRouter and some others show "Invalid API key" with the provider's error code.) If you're sure the key is right, tick **Skip API key validation** (it appears once you start typing) and save again. Your first message will show any real problem. ### Replace or delete a key With a key saved, use **Replace API key** or **Delete API key**. After a delete, that provider's models leave the model picker until you add a key again. ## Choose which models appear Each provider has a **Models** list: - The switch shows or hides a model in the model picker. The trash button removes it. - **Add New Model** lets you search the provider's models or type a **Model ID** exactly as the provider writes it. **Display Name / Alias (optional)** is the name the picker shows. Chat models get a short test message first, so a mistyped ID fails right away. - Click a model to change its alias. ## Where your key is stored Keys are encrypted with your operating system's secure storage and stay on this computer; they don't sync, so add them on each computer you use. If secure storage is locked or unavailable, the row shows an amber warning and a **Retry secure storage** button. On Linux you need a system keyring. Messages on your own key don't pass through Slashspace's servers. Usage analytics record details such as which model ran, not what you wrote. A few canvas features, like capturing web pages and reading documents, use Slashspace's servers whatever model you pick; see [Privacy and your data](/docs/privacy). ## Plans and cost API keys work on every plan. Slashspace itself needs an active plan ([Plans and billing](/docs/plans-and-billing)), but messages on your own keys never draw from the hosted wallet. ## Provider-specific options ### OpenRouter: images on your key Turn on **Use my OpenRouter key for image generation** (off by default) to bill Image mode to your OpenRouter account instead of the hosted wallet. Generation still runs through Slashspace's servers, so your OpenRouter key is sent with each image request. Without a saved key, images use the hosted wallet. See [Image generation](/docs/ai-features/image-generation). ### Groq: voice input With a Groq key saved, [voice input](/docs/ai-features/voice-input) transcribes on your own Groq account. Without one, it uses Slashspace's transcription service. ### Fal AI (BETA) You can save a Fal key and browse Fal's image models, but Image mode doesn't use Fal in this version. > **Note:** Straico is no longer supported. Keys and models from older versions were removed when you updated. ## Related - [Subscription providers](/docs/settings/subscription-providers): Claude Code, Codex and Cursor. - [Ollama (local models)](/docs/settings/ollama) - [Custom providers](/docs/settings/custom-providers): any OpenAI-compatible endpoint. # Connectors (https://www.slashspace.ai/docs/settings/connectors) Connect the services and MCP servers your Agent mode chats can use, and choose which ones are on. Connectors let the AI in [Agent mode](/docs/ai-features/agent-mode) reach the tools where your material already lives: email, documents, notes, code. With Gmail, Notion or GitHub connected, a chat on your canvas can pull a thread, a page or an issue straight into the research. You manage them in **Settings**, under **AI Settings** > **Connectors**. Everything connected there is on in every Agent mode chat unless you switch it off. To browse connectors and their tools first, see the public [connectors directory](https://www.slashspace.ai/connectors). ## Your connections This tab lists everything you've added or connected. Type in **Filter your connections** to search, or use the **All**, **Connected** and **Not connected** filters. ### Status | The row shows | Meaning | |---|---| | A green check and **Connected** | Ready to use. | | **Connecting** | A sign-in is open in your browser. The row updates when you finish. | | An amber warning and **Reconnect** | The connection failed, expired or was revoked, or a custom connector's last check failed. Wider windows show the reason beside the button. | | A key icon and **Sign in** | A custom connector that needs browser sign-in. | | **Connect** | Added but not connected yet. | ### The switch The switch decides whether a connector is offered in chats. Off hides its tools from every chat without disconnecting or deleting anything. ### Row actions The **⋯** button at the end of a row opens its actions: - **Included connectors** (from **More connectors**): **Disconnect** removes the connected account. Connect it again from Discover when you need it. - **Partner and custom connectors**: **Edit** reopens its settings, **Check connection** tests it again, and **Remove** deletes it, with its saved headers, after you confirm. ## Discover Find new connectors here, and search them with **Search connectors**. - **Aligned Partner**: ChatAcademia, featured at the top, gives your chats research tools for finding academic papers, exploring research questions and strengthening your writing. **Add connector** opens the connector form with its details filled in (**Finish setup** if you started earlier). - **Partner connectors**: remote MCP servers from partners, marked **Partner**, when any are listed. **Add** opens the connector form prefilled. - **More connectors**: a large directory of services such as Gmail, Slack, Notion and Linear, provided through [Composio](https://composio.dev). These are included connectors: Slashspace handles the sign-in and the connection. The list loads 20 at a time; **Load more** shows the next batch. ### Connecting a service 1. In **More connectors**, click **Connect** on the service's row. A sign-in page opens in your browser and the row reads **Connecting**. 2. Sign in and approve access. The browser page confirms it and offers **Return to Slashspace**. 3. Back in the app, a confirmation such as "Gmail is connected." appears and the service joins **Your connections**. Finish within two minutes, or Slashspace drops the attempt and you'll need to connect again. Included connectors need you to be signed in with an active plan, both to connect and to use in chats. The yearly subscription and Lifetime both count. Without one, you'll see "Included connectors need an active plan." with **See plans**. See [Plans and billing](/docs/plans-and-billing). ## Custom connectors To connect any MCP server yourself, click **Add** at the top of the page. The form takes a remote server's URL, or a command that **Runs on this computer**, and checks the connection before saving; [Custom MCP connectors](/docs/settings/mcp-servers) covers every field. ## Using connectors in chat Connectors work in Agent mode chats. To turn one off for a single chat, uncheck it under **Connectors** in the chat's **+** Tools menu or the `/` menu. If a tool call finds a service isn't connected, the chat shows a **Connect** card whose **Open Connectors** button takes you to it in Discover. Tool calls and approvals are covered in [Agent mode and tools](/docs/ai-features/agent-mode). ## Privacy and storage - **Included connectors**: you sign in on a page hosted by Composio, which keeps the account's access, linked to your Slashspace account. The credentials are never stored on your computer, and the connections appear on every computer you sign in on. When a chat uses one, tool calls go from your computer to Composio, which calls the service. - **Partner and custom connectors** are saved on this computer only, and Slashspace talks to the server directly. Header values and sign-in tokens are encrypted with your system's secure storage (the Keychain on macOS). Environment variables for a connector that runs on this computer are saved unencrypted. - **The AI model** receives the tool list, the calls and their results, like the rest of the conversation. - **Switches** are saved per computer. Only add connectors you trust: their tools can read, change or send data in the service they connect to. See [Privacy and your data](/docs/privacy). ## Troubleshooting - **A row shows Reconnect**: click it. An included connector gets a fresh sign-in; a custom connector is checked again, with a sign-in if needed. - **"The sign-in did not finish in time"** or **"was not completed"**: click **Connect** again and finish in your browser within two minutes. - **"Enter the headers again"**: Slashspace couldn't read the saved header values on this computer. Choose **Edit** from the row's **⋯** menu and retype them. - **Included connectors are missing**: the page shows a note when you're signed out or offline. Custom connectors still show. - **The AI ignores a connector**: check that the chat is in Agent mode, the connector reads **Connected**, its switch is on, and it's checked in the chat's **Connectors** menu. For anything else, see [Troubleshooting and support](/docs/troubleshooting). # Custom providers (https://www.slashspace.ai/docs/settings/custom-providers) Connect any OpenAI-compatible server, such as a local model or a company gateway, and use its models in your chats. A custom provider connects Slashspace to a model server that isn't in the built-in list: a local model in LM Studio or vLLM, your company's AI gateway, or a smaller hosted service. A private or fine-tuned model can then answer on the same canvas as a commercial one. Messages to a custom model go straight from your computer to that server. They don't pass through Slashspace's servers or draw from the hosted wallet. ## What OpenAI-compatible means The server accepts requests in the format of OpenAI's Chat Completions API: Slashspace sends each message to your base URL plus `/chat/completions`, with your API key as a bearer token. Most self-hosted servers and gateways support this. For Ollama or the built-in services, use [Ollama](/docs/settings/ollama) or [API providers](/docs/settings/api-providers). ## Add a provider Custom providers are included with the yearly subscription and Lifetime ([Plans and billing](/docs/plans-and-billing)). If your plan doesn't include them, the page shows a lock and an **Upgrade** button. 1. Open **Settings** > **Custom Providers** and click **Add Custom Provider**. 2. Fill in: - **ID**: a short identifier such as `lm-studio`. Use letters, numbers and dashes (a colon breaks it). It can't be changed later. - **Name**: the heading its models get in the model picker. - **API Key (Optional)**: leave empty if the server doesn't need one. Keys are encrypted on this device. - **Base URL**: the address including its version path, such as `http://localhost:1234/v1` (LM Studio) or `http://localhost:8000/v1` (vLLM). 3. Click **Add Provider**. When "Provider added" appears, close the dialog. {/* VIDEO: Add LM Studio as a custom provider, add one model, then pick it in a chat node's model picker. */} ## Add models Slashspace can't fetch a model list from a custom server, so you add each model by ID. 1. Expand the provider's row and click **Add Model**. 2. Enter the **Model ID** exactly as the server names it (most servers list IDs at the base URL plus `/models`). 3. Under **Modalities**, check **Text to Text**, then click **Add Model**. Slashspace sends a short test message and adds the model only if the server answers; otherwise the server's error appears in the dialog. The model then appears in the [model picker](/docs/ai-features/models) under the provider's name. ## Edit, turn off or delete - A provider's switch hides or shows its models in the picker. - **Edit** changes the name, key or base URL. Leave the key blank to keep it, or check **Remove saved API key**. Save with **Update Provider**. - The trash icon deletes the provider and its models, permanently. - Each model has its own switch and trash icon. Click a model to set an **Alias** or **System Prompt**. ## Limitations Slashspace can't tell what a custom model supports: - In **Agent** mode the model is offered [tools](/docs/ai-features/agent-mode). If it can't call tools, replies may fail; use **Chat** mode instead. - Connected Image nodes and Scratch pads are sent as images, which a text-only model may reject. - Only a bearer token is sent. Servers that need other headers won't work. - Image mode and Orchestrator mode don't use custom models. ## Troubleshooting A failed chat shows the error, usually the one your server sent. - **"Cannot connect to API"**: nothing answered. Check the server is running on that port and, if remote, accepts network connections. - **Not found**: the base URL probably lacks its version path, usually `/v1`. - **Unauthorized**: click **Edit** and re-enter the key. - **Model not found**: the Model ID doesn't match the server's `/models` list. - **Missing from the picker**: turn on both the provider's and the model's switch. # Custom MCP connectors (https://www.slashspace.ai/docs/settings/mcp-servers) Add any MCP server as a connector, whether it's a hosted service or a program on your computer. The Model Context Protocol (MCP) is an open standard for connecting AI apps to outside tools. An MCP server offers tools, such as searching a database or reading files, that any app speaking MCP can call. So anything that speaks MCP can become a connector, from a vendor's hosted server to a script on your own machine. Its tools then work in [Agent mode](/docs/ai-features/agent-mode) chats, next to the sources on your canvas. Ready-made connectors are covered in [Connectors](/docs/settings/connectors). > **Note:** Only add servers you trust. A connector's tools can read, change or send data, and whatever a chat passes to a tool goes to that server. ## Add a remote server 1. Open **Settings > Connectors** and click **Add**. 2. In **Add a custom connector**, enter a unique **Name** and the server's **URL**, for example `https://mcp.example.com/mcp`. 3. If the server takes an API key, click **Add header** and enter it, for example `Authorization` with `Bearer YOUR_KEY`. 4. Click **Add**. While the button reads **Checking**, Slashspace connects and lists the server's tools. It tries the Streamable HTTP transport, then SSE, and remembers whichever answers. If the check fails, **Could not connect** shows why. Fix it and try again, or click **Save anyway**. Chats only use a connector once a check passes. ### Sign in with OAuth If the server uses browser sign-in, a **Connect your account** card appears. Click **Sign in**, approve access in your browser within two minutes, and return to Slashspace, which saves and checks the connector. Or click **Save for later** and use the **Sign in** button on its row when you're ready. Your browser hands the sign-in back to Slashspace through a local address (`127.0.0.1`). Tokens refresh on their own after that. > **Tip:** Leave **Headers** empty for a server that uses sign-in. With headers present, Slashspace treats them as your credentials and doesn't offer sign-in. ## Add a server that runs on this computer Switch on **Runs on this computer** and fill in **Command**, **Arguments, one per line**, and **Environment** (click **Add variable** for each variable the server needs, such as an API key). For example, the reference filesystem server (it needs Node.js), limited to one folder: set **Command** to `npx` and the arguments to: ```text -y @modelcontextprotocol/server-filesystem /Users/you/Documents ``` On Windows, set **Command** to `cmd` and add `/c` and `npx` as the first two argument lines. Clicking **Add** runs the command right away to check it. Later, Slashspace starts it when a chat needs it and stops it after about 30 minutes unused. **The command runs with your user account's permissions**: it can reach anything you can, so scope it narrowly. It inherits only basics such as `PATH` and `HOME`, so add every variable it needs under **Environment**. ## Edit, turn off or remove On the connector's row in **Your connections**: - **The switch** hides its tools from every chat without deleting it. - **⋯ > Edit**: saved header values stay hidden, so leave one blank to keep it. A renamed header needs its value again, and a new URL needs a new sign-in. To change between remote and local, remove and re-add. - **⋯ > Check connection** reconnects and refreshes the tool list. - **⋯ > Remove** deletes it, with its saved headers and sign-in, after you confirm. An amber warning with **Reconnect** means the last check or chat use failed; the reason shows beside it. Connectors are saved in `mcp.json` under `~/Library/Application Support/Slashspace/config/` on macOS, or `%APPDATA%\Slashspace\config\` on Windows. Header values and sign-in tokens go in a separate file, encrypted with your system's secure storage. Environment values for local servers stay in `mcp.json` as plain text. ## Add to Slashspace links Server authors can publish an **Add to Slashspace** link. Opening one launches Slashspace with **Add a custom connector** prefilled with a name and HTTPS URL (never keys or local commands); you review it and click **Add**. To make one, see [Add to Slashspace button](/docs/settings/add-to-slashspace). ## Troubleshooting - **"Timed out after 15 seconds"**: check the URL. A first `npx` run downloads the package and can outlast the check, so run the command once in a terminal, then try again. - **`ENOENT`**: Slashspace can't find the program. Use its full path (`which npx` in Terminal prints it), or the `cmd` form on Windows. - **401 with headers**: the key is wrong or expired. Edit and retype it. - **Sign-in fails**: start again from Slashspace rather than reusing an old browser tab. "Encrypted storage is required to sign in to a connector" means your system has no secure storage. - **Stops mid-chat**: the chat warns it is unavailable and skips it until a check passes. Fix the cause, then click **Reconnect** (or **Sign in**, if its sign-in expired). - **Logs**: **Help > Open Log Folder** (`⌘⇧L`, or `Ctrl+Shift+L` on Windows and Linux) opens the folder with Slashspace's log, which records each failed check. A local server's own errors aren't captured there; run its command in a terminal to see them. More help: [Troubleshooting and support](/docs/troubleshooting). # Ollama (local models) (https://www.slashspace.ai/docs/settings/ollama) Run open models such as Llama, Qwen and Gemma on your own computer, with no per-message cost. Ollama is a free app that runs open models on your computer. When a chat node uses a local model, its conversation and connected context go to Ollama on your machine instead of a cloud provider. Long research sessions cost nothing per message, conversations stay on your computer, and chats work offline. The trade-off: your hardware sets the speed and the quality ceiling. ## Set up Ollama 1. Open **Settings** and click **Ollama** under **AI Settings**. If Slashspace can't reach Ollama, the page shows **Please Start / Install Ollama**. 2. Click **Install Ollama** to open Ollama's download page. Install Ollama and start it (open the app, or run `ollama serve`). 3. Leave the Ollama settings page and open it again. Once Slashspace finds Ollama, the page shows **Installed** and Ollama's version. ## Download a model Models you've downloaded into Ollama appear in Slashspace automatically. To get one, type its name (for example `llama3.2`) into the field on the Ollama settings page and click **Download Model**. **Search Models on Ollama** opens Ollama's model library. You can also run `ollama pull llama3.2` in a terminal. Models loaded in memory show **Running**. **Delete** removes a model from your computer. Choose a model whose download size sits well under your computer's memory (or your graphics card's). Small models (a few billion parameters) suit most laptops; bigger ones answer better but slower. {/* VIDEO: Download a model on the Ollama settings page, then pick it in a chat node's model picker and send a message. */} ## Use a local model Open the model picker in a chat node and choose a model from the Ollama group. To start new chats on it, make it your [default model](/docs/ai-features/models#default-models). - **Cost:** local models don't draw from the hosted wallet. Like every chat, they need an active plan: the yearly subscription or Lifetime. See [Plans and billing](/docs/plans-and-billing). - **Agent mode:** new chats start in [Agent mode](/docs/ai-features/agent-mode), which gives the model tools, so only models that support tool calling work there. For other models, switch the chat to **Chat** mode (`⌘2`, or `Ctrl+2` on Windows and Linux). Ollama's library tags models that support tools. - **Images:** for connected Image nodes or a Scratch pad, pick a model tagged vision. - **Orchestrator** and **Image** modes don't use Ollama. ## What still uses the network - **Sign-in:** Slashspace checks your sign-in before each chat and renews it while you're online. After a couple of days offline, local chats can stop with a sign-in error. - **Capturing sources:** [Web](/docs/nodes/web-node) and [Post](/docs/nodes/post-node) capture and [Document](/docs/nodes/document-node) processing run on hosted services. Once captured, the text is saved with your canvas. - **Chat titles:** by default, a hosted service names each new chat from its first message. - **Connectors** reach their own services, and voice input transcribes in the cloud. - **Usage analytics** (not your messages) go to PostHog. See [Privacy and your data](/docs/privacy). ## Connection settings The gear next to Ollama's version (or **Configure** when Ollama isn't found) opens **Ollama Configuration**. **Hostname** is where Slashspace looks for Ollama, `http://localhost:11434` by default; **use the default** restores it. > **Note:** Chats always go to Ollama on this computer at the default address, whatever the hostname says. Keep Ollama running locally on port 11434. ## Troubleshooting - **Ollama isn't detected.** Open `http://localhost:11434` in a browser; it should say "Ollama is running". Check that **Hostname** is the default, then reopen the settings page. - **Models are missing from the picker.** If you started Ollama after Slashspace, they can take up to five minutes to appear. Restarting Slashspace shows them right away. - **Download Model does nothing.** An unknown name fails silently. Copy the exact name from Ollama's library, or run `ollama pull` in a terminal to see the error. - **Replies are slow.** The first reply after a pause waits while Ollama loads the model. Try a smaller model and connect only the nodes a chat needs. - **Answers miss part of a long context.** Ollama cuts off anything beyond its context length, which you can raise in Ollama's settings. More fixes in [Troubleshooting and support](/docs/troubleshooting). # Agent skills (https://www.slashspace.ai/docs/settings/skills) Install ready-made skills from the skills.sh registry, add them to any chat with /, and write your own. A skill is a packaged set of instructions for one kind of task: a research method, a writing style, a review checklist. Add it to a message and the AI follows it, so you don't retype the same long prompt in every branch. You don't have to write them yourself. The **Agent Skills** page installs skills published on [skills.sh](https://skills.sh), a public registry, and makes them available to Claude Code too. ## Browse the registry Open **Settings** and choose **Agent Skills** under **Templates** (it needs an internet connection). Skills appear in skills.sh leaderboard order: | Column | Shows | |---|---| | **#** | Leaderboard rank | | **Skill** | Name, a **Verified** badge for official publishers, the source GitHub repository, and a short description | | **Checks** | A shield summarizing third-party security audits. Hover it for each result | | **8W trend** | Weekly installs over the last eight weeks | | **Installs** | Total installs on skills.sh | Click a row for the full description, each security check and a **View on skills.sh** link. Type at least two characters in **Search skills** to search the whole registry, beyond the leaderboard. There's no sort option. ## Install, update and remove Click **Install**. The skill is saved to `~/.agents/skills` in your home folder, the same place the skills.sh installer uses, and linked into `~/.claude/skills` so Claude Code finds it. The skill appears in the `/` menu right away. To remove a skill, click **Installed** on its row. It's deleted immediately, without confirmation, along with its link. There's no update button: to get a newer version, remove the skill and install it again, which discards any edits you made. > **Note:** A chat that used a skill you later removed (and any chat that reads it through a connection) shows "Missing slash command reference" when you send, until you install the skill again. ## Use a skill Type `/` in a chat, open **Skills**, and pick one (or type part of its name). It becomes a chip in your message, and when you send, Slashspace inserts the skill's instructions at that spot. [Slash commands, skills and prompts](/docs/ai-features/slash-commands) covers the rest of the menu. Skills work with every model, including Claude Code and Codex. Agent mode never loads a skill on its own: you add it with `/`. Skills don't take effect in Orchestrator mode. Slashspace sends only the text of the skill's `SKILL.md` file. Registry skills that rely on bundled scripts or reference files may only partly work. ## Write your own skill Create a folder in `~/.agents/skills` containing a `SKILL.md` file that starts with a `---` line: ```md --- name: Source comparison description: Compare sources in a table of claims, evidence and gaps --- When I share several sources, build a table with one row per source... ``` `name` is the label in the `/` menu (the folder name if you leave it out), and `description` shows under it. Everything after the frontmatter is what the AI receives, up to 120,000 characters. Slashspace rereads the file on every send, so edits apply to your next message. Folders nested inside `~/.agents/skills` work; symlinked folders and `~/.claude/skills` are not read. > **Note:** The Agent Skills page recognizes installed skills by folder name. If your folder shares a name with a registry skill, that row shows **Installed**, and clicking it deletes your folder. ## Review before you install Skills are instructions written by other people, and the AI follows them. In Agent mode they can steer how the AI uses your connectors, and Claude Code can run scripts bundled with a skill. Before installing, check the security results and the **Verified** badge, and read the skill on skills.sh. Passing checks are a signal, not a guarantee. ## Related - [Slash commands, skills and prompts](/docs/ai-features/slash-commands) - [Agent mode and tools](/docs/ai-features/agent-mode) # Subscription providers (https://www.slashspace.ai/docs/settings/subscription-providers) Use the Claude or ChatGPT subscription you already pay for on the canvas through Claude Code and Codex, or connect Cursor with an API key. If you already pay for Claude or ChatGPT, you can use that subscription on the canvas instead of paying again. Slashspace drives the official **Claude Code** and **Codex** command-line tools (CLIs) on your computer, signed in with your own account. **Cursor** connects with a Cursor API key instead. These models are agents: they work in a folder you choose, reading files and, if you allow it, changing them. That helps when your research lives in files as well as on the canvas. ## How it works You install a CLI once and sign in to it with the account you use in the browser. When a chat node that uses one of its models sends a message, Slashspace starts the CLI on your computer, passes it the message and the chat's context, and streams the reply into the node. Usage counts against your Claude or ChatGPT plan, with no API key to paste. Setup lives in **Settings** (bottom of the sidebar) > **Subscription Providers**, under **AI Settings**: - The **Slashspace** card at the top is for Slashspace's own hosted models, paid from your optional hosted wallet. It needs no setup. **Manage hosted wallet** opens your account page on the web; see [Plans and billing](/docs/plans-and-billing). - Below it, **Cursor**, **Claude Code** and **Codex** each have a row you can expand. A row's check mark turns green once that provider is connected. {/* VIDEO: Turning on Claude Code in Subscription Providers, the CLI Status card showing a green check, then picking a Claude Code model and a workspace folder in a chat node. */} ## Open a terminal You install and sign in once, in a terminal. On macOS, press `⌘Space`, type Terminal and press Return. On Windows, press the Start key, type PowerShell and press Enter. On Linux, open your terminal app. Paste a command, press Enter and wait for it to finish. When Slashspace can't find a CLI, its **CLI Status** card shows these same commands for your system, with copy buttons. ## Claude Code Claude Code uses your Claude subscription. ### Install Claude Code macOS or Linux: ```bash curl -fsSL https://claude.ai/install.sh | bash ``` Windows (PowerShell): ```powershell irm https://claude.ai/install.ps1 | iex ``` Windows (Command Prompt): ```bash curl -fsSL https://claude.ai/install.cmd -o install.cmd && install.cmd && del install.cmd ``` With npm, if you already use Node.js: ```bash npm install -g @anthropic-ai/claude-code ``` ### Sign in to Claude Code ```bash claude ``` Follow the prompts and log in with the Claude account that has your subscription, not an API key. ### Turn on Claude Code 1. In **Subscription Providers**, expand **Claude Code** and turn on its switch. Slashspace loads the model list from the CLI. 2. Check the **CLI Status** card. A green check with a version number and a file path means you're ready. If it says the CLI wasn't found, quit and reopen Slashspace (it reads your program paths at startup), then click **Re-check**. ### Claude Code options - **Custom CLI path**: leave it empty for automatic detection (see [Troubleshooting](#troubleshooting)). - **Permission mode**: how much Claude Code may do without asking you. - **Default**: asks for your approval in the chat before actions that need permission. - **Accept edits**: edits files in the workspace without asking. Other actions still ask. - **Plan**: read-only. It plans and explains but changes nothing. - **Bypass permissions**: runs everything without asking. Use with care. - **Max agent turns**: an optional cap (1 to 100) on rounds of tool use per reply. Empty means no limit. At the cap, the chat tells you; send "continue" to carry on. - **Refresh Models**: reloads the model list from the CLI, for example after your plan changes. If the CLI can't be reached, a default set loads instead. ## Codex Codex uses your ChatGPT subscription. ### Install Codex macOS or Linux: ```bash curl -fsSL https://chatgpt.com/codex/install.sh | sh ``` Windows (PowerShell): ```powershell powershell -ExecutionPolicy ByPass -c "irm https://chatgpt.com/codex/install.ps1 | iex" ``` With npm, if you already use Node.js: ```bash npm install -g @openai/codex ``` ### Sign in to Codex ```bash codex login ``` This opens your browser. Sign in with the ChatGPT account that holds your subscription, not an API key. ### Turn on Codex Expand **Codex**, turn on its switch, and check the **CLI Status** card as you would for Claude Code. ### Codex options - **Custom CLI path**: same as for Claude Code. - **Approval policy**: when Codex stops to ask you in the chat. - **On request** (default): Codex decides which commands need your approval. - **Untrusted**: every command asks first. - **On failure**: commands run in the sandbox, and you're asked only when one fails. - **Never**: never asks. Everything runs within the sandbox. - **Sandbox mode**: how much of your computer Codex can touch. - **Workspace write** (default): can change files inside the agent workspace. - **Read-only**: can read the workspace but not change anything. - **Full access**: no sandbox, so Codex can touch anything on your computer. Use with care. - **Refresh Models**: same as for Claude Code. ## Cursor Cursor needs no CLI. It connects with a Cursor API key, and Cursor, not Slashspace, bills you for what you use. 1. Expand **Cursor** and click **Get API key** to open your Cursor dashboard and create a key. 2. Paste it into the **Add API key** field and click the check button. Slashspace tests the key before saving it. 3. Click **Refresh Models** to load Cursor's current models. Cursor's agent also runs on your computer in the agent workspace, but Slashspace has no permission or sandbox settings for it. ## Agent workspace The agent workspace is the folder these models read and, depending on your settings, change. Choose one before your first message. - **For one chat:** when a chat node uses one of these models, a folder button appears right below its message box. Hover it for the full path; click it to choose a folder. - **For a whole canvas:** click the gear in the canvas header to open **Canvas settings**, then under **Agent workspace** click **Choose folder...**. Every agent chat on the canvas uses that folder unless you picked one for the chat itself. The X next to the folder name (**Clear workspace**) removes it. > **Note:** These agents can change and delete files in the workspace, and **Accept edits**, **Bypass permissions** and **Full access** let them do it without asking. Pick a folder that holds only what they should work on, keep a backup, and start with **Plan** (Claude Code) or **Read-only** (Codex) until you trust the results. ## Using these models in chats Once a provider is connected, its models appear in every chat node's model dropdown, in **Claude Code**, **Codex** and **Cursor** groups below the **Slashspace** group. You can also make one your default; see [Choosing a model](/docs/ai-features/models). They never draw from the hosted wallet, so they keep working at a zero balance. Hosted tools such as image generation still charge the wallet. A few things work differently from Slashspace's own models: - They use their own tools, such as reading files and running commands. You see the tool calls in the chat, where Claude Code and Codex ask for approval according to the settings above. See [Agent mode and tools](/docs/ai-features/agent-mode). - In Agent mode, the connectors turned on for the chat come along, and these models can hand work to [sub-agents](/docs/ai-features/sub-agents). - They can't suggest switching the chat's mode or model. With no incoming connections they get only a short summary of the rest of the canvas, so [connect](/docs/canvas/connections) the nodes you want them to read in full. ## Troubleshooting ### Slashspace can't find the CLI Check that the install finished without errors, quit and reopen Slashspace, and click **Re-check**. If you installed the CLI somewhere unusual (for example with a version manager), find it with `which claude` or `which codex` in a terminal (on Windows, `where.exe claude` or `where.exe codex`). Paste the path into **Custom CLI path** and click **Save**. Empty the field and save to return to automatic detection. If the card then says "No executable found", or on Windows mentions a "launcher script", the saved path isn't the program itself. Point **Custom CLI path** at the program, or reinstall with the PowerShell command. ### The CLI isn't signed in **CLI Status** checks that the CLI runs, not that you're signed in. If you aren't, **Refresh Models** falls back to the default list with a note that it couldn't fetch models from the CLI, and messages fail with a sign-in error. Run `claude` or `codex login` in a terminal again. Slashspace uses whichever account the CLI is signed into. To switch accounts, sign out and back in from the terminal. ### Usage is billed to an API key Slashspace passes your computer's environment settings to the CLI. If `ANTHROPIC_API_KEY` or `OPENAI_API_KEY` is set, or you first signed the CLI in with an API key, it may bill that key instead of your subscription. Run the CLI in a terminal to see which account it uses. For other errors, see [Troubleshooting and support](/docs/troubleshooting). ## Related - [Choosing a model](/docs/ai-features/models) - [Agent mode and tools](/docs/ai-features/agent-mode) - [API providers](/docs/settings/api-providers), to pay per use with your own keys instead # Troubleshooting and support (https://www.slashspace.ai/docs/troubleshooting) Find the message or symptom you're seeing, fix it, and reach support when you're still stuck. Find what you're seeing below, then follow the fix. If nothing matches, [get help](#get-help). ## Chat errors Chat errors appear as a banner at the bottom of the conversation. ### "Your hosted wallet has insufficient funds" The hosted wallet, which pays for Slashspace's hosted models (including **Auto**) and hosted tools such as image generation, is empty. Your plan is unchanged. Either: - Click **Add funds** to top up on your account page. See [Plans and billing](/docs/plans-and-billing). - Switch to a model that doesn't use the wallet: your own API key, Claude Code, Codex, Cursor or Ollama. See [Choosing a model](/docs/ai-features/models). Generated images show this as **Out of usage balance**. Switching models still works. ### "Your trial has ended" and other plan messages A banner with **Upgrade**, such as "Your trial has ended. Your canvases are still here. Subscribe to keep using Slashspace.", means your plan has ended or doesn't cover the request. Click **Upgrade** for options. If you've already paid, click **Refresh account** in **Settings** > **Account**. ### "Sign in to Slashspace to use chat." Your session ended (you may also see a **Session expired** banner). In **Settings** > **Account**, click **Sign out**, then sign in again. ### A red error with Retry Click **Retry**: brief outages and rate limits are common. If it keeps failing, try another model to see whether one provider is down. For "No API key found for provider" or an invalid-key error, add or replace the key in [API providers](/docs/settings/api-providers). ### "Missing slash command reference" A skill or prompt chip points to a file that was moved, renamed or deleted. Remove the chip and pick it again from the `/` menu. If the error persists, the chip is in an earlier message: put the file back. See [Slash commands](/docs/ai-features/slash-commands#if-something-goes-wrong). ## Sign-in problems ### The browser signed in, but the app didn't On the browser page, click **Open Slashspace**. If nothing happens, click **Copy login token** there, then in the app click **Paste login token**, paste, and click **Continue sign-in**. "Sign-in session expired or invalid. Please try again." means the token came from an older attempt. Click **Sign in** again and use the newest browser tab. ### "All your device slots are in use" Click **Remove a device in account settings, then sign in again.** Remove an old computer on the page that opens, then sign in. Details: [Account and devices](/docs/settings/account-management#when-all-your-device-slots-are-in-use). ### "Update Slashspace to sign in" Your version is too old for your license. Click **Check for updates**, or **Download from the website** and install over the old version. Reinstalling doesn't touch your canvases. ### The plan screen appears, but you already pay Click **Already subscribed? Refresh account**. If that doesn't help, click **Sign out** and sign in with the email you bought Slashspace with. ## Sources that don't load ### A Web node shows nothing A link pasted onto the canvas waits for you: click **Go** in the node. A red message usually means a login wall or paywall. Copy the text into a [Text node](/docs/nodes/text-node), or save the page as a PDF and add it as a [Document node](/docs/nodes/document-node). More causes: [Web node](/docs/nodes/web-node#when-a-capture-fails). ### A pasted link became a Text node Links containing `#`, `@`, `~`, `+`, commas, parentheses or a port number (such as `:8080`) aren't recognized. Press `W`, click the canvas, and paste the link into the "Enter URL" field. ### A Post node transcript shows Failed Click **Failed** to read the reason, then **Retry**. "Active desktop access is required to fetch media transcripts" means your plan isn't active. Private posts and videos with no speech can't be transcribed. See [Post node](/docs/nodes/post-node#when-a-transcript-fails). ### A document is stuck or Failed - **Failed** or **Timed Out**: click the circular-arrow button next to the status. - Never reaches **Ready**: delete the node and add the file again. - A Word or PowerPoint file keeps failing: export it as a PDF and add that. See [Document node](/docs/nodes/document-node#if-processing-fails). ### An image is blank Slashspace only loads images inside your home folder, so files on another drive (such as `D:\` on Windows) stay blank and aren't sent to chats. Copy the file into your home folder and add it again. Added images are linked, not copied, so moving or deleting the original breaks the node. Generated images can go missing after you rename or move their canvas; renaming or moving it back restores them. ## Models ### "Please Start / Install Ollama" Slashspace can't reach Ollama. Start the Ollama app and reopen **Settings** > **Ollama**. If it runs on another address, click **Configure**, set the **Hostname** (default `http://localhost:11434`) and click **Save Configuration**. Models appear in the picker once you've downloaded one. See [Ollama](/docs/settings/ollama). ### "Claude Code CLI not found" or "Codex CLI not found" Install the CLI, restart Slashspace and click **Re-check**, or set **Custom CLI path**. If messages fail with a sign-in error, sign in again from a terminal. See [Subscription providers](/docs/settings/subscription-providers#slashspace-cant-find-the-cli). ### "Could not validate the API key" Check that you copied the whole key, that billing is set up with the provider, and that you're online. If the key is right, tick **Skip API key validation** and save. See [API providers](/docs/settings/api-providers#if-the-key-is-rejected). ## Context surprises ### A chat ignores its connections The chat is in **Isolated context** (its left handle stays gray). Isolated chats ignore connections, and the mode switch is hidden while any exist. To fix it: 1. Select each line coming into the chat and press `Backspace` or `Delete`. 2. Hover the left handle and click the icon to switch to **Summarized context**. 3. Reconnect your sources. If only a document or video is missing, wait for its node to show **Ready**. See [Connections and context](/docs/canvas/connections#context-modes). ### A branch sees nothing from its parent A branch copies its parent's context mode, so a branch of an isolated chat ignores its parent too. Switch the parent to **Summarized context** before branching, or fix the branch with the steps above. See [Branching](/docs/canvas/branching#what-a-branch-sees). ## Shortcuts on Windows and Linux Two canvas shortcuts are Mac only: | Mac | Windows and Linux | |---|---| | `⌘B` on the canvas (empty branch) | Double-click the node's right handle, or press `Ctrl+B` while typing in a chat | | `⌘C` to copy nodes | Click **Duplicate** in the node's header (**Duplicate Node** on chats) | Redo is `Ctrl+Y` on Windows and `Ctrl+Shift+Z` on Linux. Search opens with `Ctrl+K`, even though its tooltip shows `⌘K`. See [Keyboard shortcuts](/docs/canvas/keyboard-shortcuts). ## App updates Slashspace checks for updates when it starts. When one is available, click **Update now**, then **Restart** once it downloads. To check yourself, open **Settings** and click the version number at the bottom of the sidebar. If an update fails, install the latest version from [slashspace.ai/download](https://www.slashspace.ai/download) over the old one. If your plan includes new versions only until a date, later versions aren't offered: you see "Version X is outside your updates window". The version you have keeps working. See [Plans and billing](/docs/plans-and-billing). ## Logs and bug reports To find the log file (`main.log`), choose **Help** > **Open Log Folder**, or press `⌘⇧L` (`Ctrl+Shift+L` on Windows and Linux). A useful report says what you did, what you expected and what happened, with the exact error or a screenshot, your Slashspace version (bottom of the **Settings** sidebar), your operating system, and the chat's mode and model. **Bug / Feedback** attaches your log automatically, with email addresses and file paths removed. A log you send yourself isn't cleaned, so check it first. ## Get help - **In the app:** right-click the canvas and choose **Bug / Feedback**, or go to **Settings** > **Troubleshoot** > **Bug / Feedback**. Add screenshots or screen recordings if they help. You need to be signed in. - **Email:** [support@slashspace.ai](mailto:support@slashspace.ai), including for sign-in problems. - **Discord:** [links.rabbitholes.ai/discord](https://links.rabbitholes.ai/discord). # Why a canvas (https://www.slashspace.ai/docs/why-a-canvas) Where a single chat thread breaks down for long, exploratory work, and what changes when every conversation and source is a node on a canvas. Slashspace is an AI infinite canvas: a workbench for long, exploratory work with AI. Research, learning and writing rarely move in a straight line. You branch, backtrack, compare, and connect ideas that live far apart. A chat thread holds one line of thought at a time, and only its last few messages fit on screen. A canvas holds the whole shape of the work: chats, branches and sources sit side by side, so you can see far more at once, keep more of the problem in your head and spot how the pieces fit together. You also decide what the AI reads at every step. ## Where a chat thread breaks down A single scrolling chat is fine for a quick question. It starts to fail when the work runs for hours or weeks: - **You can only see the last few messages.** Everything earlier scrolls out of view, so you carry the rest of the picture in your head or scroll back to find it. The bigger the problem, the more slips away. - **Tangents have nowhere to go.** A side question lands in the main thread and every later answer reads around it. Skip it and the idea is gone. - **You lose sight of what the model is reading.** The conversation grows until you can't see, or control, which parts of it shape the next answer. - **Your sources live somewhere else.** Papers, pages and videos sit in other tabs and get pasted in piece by piece. - **Comparing means copy-paste.** Trying a second approach or model means a fresh chat and pasting everything again. - **Weeks later you have a transcript, not a map.** The reasoning is in there somewhere, under hundreds of messages. ## What the canvas changes **You can see the whole problem at once.** Chats, branches and sources stay laid out where you put them, so a glance takes in what a long scroll hides. Zoom out to see the shape of the work, or open the mini map, then zoom back in to read one part. See [Navigating the canvas](/docs/canvas). **Everything is a node.** Conversations are [Chat nodes](/docs/nodes/chat-node) and notes are [Text nodes](/docs/nodes/text-node). A web page, a YouTube lecture, a PDF and an image each get their own node ([Web](/docs/nodes/web-node), [Post](/docs/nodes/post-node), [Document](/docs/nodes/document-node), [Image](/docs/nodes/image-node)), and a [Scratch pad](/docs/nodes/scratch-pad-node) holds a quick sketch. **Connections decide what each conversation sees.** Connect a source to a chat and the chat reads it. Chains carry through: a chat reads everything connected upstream of it, and nothing else. A chat with nothing connected gets a summary of the whole canvas instead, or nothing at all when you switch it to **Isolated context**. See [Connections and context](/docs/canvas/connections). **Any answer can fork into a branch.** Send your next question with **Submit in new branch** and it opens in a new chat node beside the original. The branch picks up the conversation so far, plus everything connected to it, and the original thread stays as it was. See [Branching](/docs/canvas/branching). {/* VIDEO: forking one answer into two branches, then switching one branch to a different model and asking the same follow-up in both. */} **Sources sit next to the conversations that use them.** Paste a link, drop a PDF onto the canvas, or type `@` in a chat to mention a node and connect it. **Different models work side by side.** Each chat node has its own model, so you can put the same question to Claude, GPT, Gemini and a local model and read the answers next to each other. Use Slashspace's hosted models, your Claude or ChatGPT subscription ([Subscription providers](/docs/settings/subscription-providers)), your own [API keys](/docs/settings/api-providers), or local models through [Ollama](/docs/settings/ollama). See [Choosing a model](/docs/ai-features/models). **The AI can take on more of the work.** In [Agent mode](/docs/ai-features/agent-mode), a chat can use your [connectors](/docs/settings/connectors) and split a job across [sub-agents](/docs/ai-features/sub-agents), each working in its own chat node. In [Orchestrator mode](/docs/ai-features/orchestrator), you describe a goal and the AI builds the canvas: it creates and connects chat, text, web and post nodes, and hands research tasks to sub-agents. **Your canvases are files you own.** Each canvas is saved in a [folder on your computer](/docs/canvas/canvases-and-folders). Messages go to the model provider you pick, and some features, such as reading web pages and documents or Orchestrator mode, process content on Slashspace's servers. [Privacy and your data](/docs/privacy) covers what leaves your computer. ## What a session looks like ### Researching a question across papers You want to know whether a new method holds up. Drop four PDFs onto the canvas, paste links to two critical blog posts, connect them all to a chat, and ask where the sources disagree. When an answer cites a figure you doubt, select that sentence and type your question in the "Ask about this…" box: it opens in its own branch and the main thread stays on track. Later, connect a second chat to the two critiques only, for a reading the original papers don't shape. ### Learning a subject from lectures Paste the links to a YouTube lecture series and each becomes a Post node with its transcript. Put each week's lectures in a [group](/docs/nodes/group-node) and connect it to a chat that quizzes you. When a concept doesn't land, branch off for a slower explanation, or sketch your understanding in a connected Scratch pad and ask whether you've got it right. Next week, the canvas shows what you covered and where you got stuck. ### Drafting a long article Connect your interview transcripts, notes and reference pages to a chat where you work out the outline. Once it holds, fork three branches, one per angle you're weighing, and ask each for an opening. Every branch reads the same outline and sources, so the drafts differ only in angle. Copy the strongest opening, paste it onto the canvas as a Text node, and connect it to the chat where you write the rest. ### Weighing a decision with several models You're choosing between three vendors. Add each pricing page as a Web node, write your constraints in a Text node, and group them. Connect the group to three chats on different models: Claude through your subscription, GPT on your own OpenAI key, a local model through Ollama. Ask all three the same question, then connect them to a fourth chat and ask where they agree. The reasoning behind the decision stays on the canvas. ## When a plain chat is enough Not every question needs a canvas. A quick fact, a one-off rewrite or a short exchange you won't revisit works fine in one thread, and in Slashspace that's a single Chat node. The canvas asks a little more of you than a chat box, because you decide what connects to what. That pays off when the work spans many sources, several lines of thought, or more than one sitting: more than you can keep in view in one scroll. ## Next - [Quickstart](/docs): install Slashspace, ask your first question, fork your first branch and connect your first source. - [Core concepts](/docs/core-concepts): the mental model behind nodes, connections, branches and chat modes. --- # Blog Posts ## 10 Ways AI Mind Mapping Can Supercharge Your Brainstorming Sessions (https://www.slashspace.ai/blog/10-ways-ai-mind-mapping-supercharge-brainstorming) Published: 2026-03-31 Traditional brainstorming hits walls. AI mind mapping changes this entirely — here are ten specific ways it transforms brainstorming from a struggle into a supercharged creative process. Traditional brainstorming hits walls. You know the feeling—staring at a blank whiteboard, ideas trickling out in single file, conversations getting stuck in loops. The problem isn't lack of creativity; it's the linear nature of how we typically capture and explore ideas. AI mind mapping changes this entirely. Instead of forcing thoughts into neat rows or bullet points, you can create sprawling, interconnected webs of concepts that branch and evolve naturally. The AI doesn't just passively record your ideas—it actively participates, suggesting connections you missed and pushing your thinking in unexpected directions. Here are ten specific ways AI-powered mind mapping transforms brainstorming from a struggle into a supercharged creative process. ## 1. Generate Unexpected Connections Between Distant Ideas Your brain excels at making connections, but it tends to follow familiar pathways. AI mind mapping tools break these patterns by identifying relationships between concepts that seem completely unrelated. When you're mapping out ideas for a new product launch, the AI might connect your thoughts about "customer onboarding" with a seemingly unrelated branch about "seasonal marketing." Suddenly, you realize that timing new user tutorials around holiday shopping patterns could dramatically improve engagement. This happens because AI processes your entire idea map simultaneously, not sequentially. While you're focused on one branch of thinking, the AI maintains awareness of every other concept you've explored, constantly scanning for non-obvious links. The result? Breakthrough insights that emerge from the intersections between different domains of your thinking. ## 2. Expand Single Ideas into Rich Concept Networks One word or phrase can explode into dozens of related concepts when AI gets involved. Start with "customer feedback" and watch it branch into sentiment analysis, survey design, social listening, review platforms, support ticket mining, and user interview frameworks. This expansion isn't random—it's contextually aware. The AI considers your industry, the specific project you're working on, and the other ideas already in your map. If you're brainstorming marketing strategies for a B2B software company, "customer feedback" will expand differently than it would for a restaurant chain. The spatial nature of mind mapping makes these expansions particularly powerful. Instead of overwhelming you with a long list, the AI creates visual clusters that help you see patterns and priorities at a glance. ## 3. Switch Perspectives Mid-Stream Without Losing Momentum Traditional brainstorming sessions often get derailed when someone suggests looking at the problem from a different angle. You have to stop, reset, and start over with the new perspective. AI mind mapping lets you maintain multiple viewpoints simultaneously. Create one branch exploring your challenge from a customer's perspective, another from a competitor's angle, and a third from an investor's viewpoint. The AI helps populate each branch with relevant considerations while maintaining the connections between them. This multi-perspective approach reveals tensions and opportunities that single-angle thinking misses. You might discover that what looks like a feature request from customers actually represents a competitive threat that requires a completely different strategic response. ## 4. Transform Vague Concepts into Actionable Frameworks "We need better user engagement" is the kind of vague starting point that kills momentum in traditional brainstorming. AI mind mapping takes these fuzzy concepts and systematically breaks them down into concrete, workable elements. The AI might expand "user engagement" into specific metrics (session duration, feature adoption, return visits), then branch each metric into measurement methods, improvement tactics, and success criteria. What started as a vague goal becomes a structured framework for action. This transformation happens iteratively. As you interact with the AI's suggestions, refining and redirecting them, the framework becomes increasingly tailored to your specific situation and constraints. ## 5. Maintain Context Across Complex, Multi-Threaded Discussions Team brainstorming sessions often fragment into multiple simultaneous conversations. Someone starts exploring pricing strategies while another person dives into technical implementation details. Valuable insights get lost in the chaos. AI mind mapping platforms like RabbitHoles AI solve this by creating separate but connected conversation threads. Each team member can pursue their line of thinking without interrupting others, while the AI maintains awareness of all parallel discussions. When connections emerge between different threads—like realizing that a technical constraint actually supports a specific pricing model—the AI can surface these insights for the whole team. Context doesn't get lost; it gets amplified. ## 6. Integrate External Knowledge Without Breaking Flow The best brainstorming sessions draw on diverse sources of information, but stopping to research facts or find references kills creative momentum. AI mind mapping tools can pull in external context without disrupting the flow of ideas. Upload relevant documents, paste in URLs, or reference existing knowledge bases, and the AI weaves this information into your brainstorming naturally. Discussing market positioning? The AI can reference competitor analysis documents you uploaded earlier. Exploring technical solutions? It can draw on engineering specs or industry best practices. This integration isn't just about access—it's about timing. The AI surfaces external information precisely when it becomes relevant to your current line of thinking, not as a separate research phase. ## 7. Capture and Organize Rapid-Fire Idea Generation High-energy brainstorming produces ideas faster than anyone can organize them properly. The good stuff gets buried under the volume, and promising directions get forgotten in the rush. AI mind mapping handles the organizational load automatically. As ideas pour out, the AI clusters related concepts, identifies themes, and maintains the relationships between different branches of thinking. You stay in creative flow while the AI handles the cognitive overhead of structure and organization. Later, when you're ready to evaluate and prioritize, the AI can help surface the most promising clusters or identify branches that didn't get fully explored during the initial burst of creativity. ## 8. Break Through Mental Blocks with Intelligent Prompting Every brainstorming session hits walls where ideas stop flowing. Traditional techniques like "what if" questions or random word association can help, but they're hit-or-miss. AI mind mapping provides more sophisticated block-breaking techniques. The AI analyzes your existing idea map to identify unexplored angles, underexamined assumptions, or missing stakeholder perspectives. Instead of generic prompts, you get targeted questions that push your specific thinking forward. "You've explored customer acquisition thoroughly, but what about customer retention in the context of your pricing strategy?" This kind of contextual prompting opens new avenues that feel natural and relevant rather than forced. ## 9. Evolve Ideas Through Multiple AI Model Perspectives Different AI models have different strengths and thinking patterns. One might excel at creative association while another provides more analytical structure. AI mind mapping platforms that support multiple models let you leverage these different capabilities within a single brainstorming session. Start with a creative model to generate wild possibilities, then switch to an analytical model to evaluate feasibility and identify implementation challenges. Each model adds its own layer of insight while building on the work of the others. This model-switching capability means you're not locked into one type of thinking. Your brainstorming can be simultaneously creative and practical, visionary and grounded, without losing coherence. ## 10. Create Persistent Knowledge Networks for Future Sessions Traditional brainstorming sessions end when people leave the room. The whiteboard gets erased, notes get scattered, and insights get forgotten. AI mind mapping creates persistent knowledge networks that grow more valuable over time. Each brainstorming session adds to your organizational knowledge base. Patterns emerge across multiple sessions, successful strategies get documented and refined, and failed approaches get marked to avoid repetition. The AI learns from your team's thinking patterns and becomes increasingly effective at supporting your specific creative processes. This persistence transforms brainstorming from isolated events into an ongoing organizational capability. Each session builds on previous insights, creating compound returns on your creative investment. ## The Spatial Advantage What makes AI mind mapping particularly powerful for brainstorming is the spatial element. Ideas aren't trapped in linear lists or sequential slides—they exist in a flexible, visual space where proximity indicates relationship and distance suggests independence. This spatial organization mirrors how your brain actually works. Related concepts cluster together naturally, while unexpected connections span across the map like bridges between islands of thought. The AI helps maintain and enhance this spatial logic, ensuring that your idea landscape remains navigable even as it grows complex. ## Beyond Traditional Brainstorming AI mind mapping doesn't just improve traditional brainstorming—it enables entirely new forms of creative collaboration. Teams can work asynchronously on the same idea space, with the AI maintaining coherence across different time zones and work sessions. Individual contributors can explore personal branches without disrupting team flow, then merge insights back into the collective thinking. The result is brainstorming that's more inclusive, more thorough, and more productive than what's possible with conventional approaches. Ready to transform your brainstorming sessions from linear struggles into dynamic, AI-powered explorations? The tools exist today to make this shift, turning creative collaboration into a genuine competitive advantage. Learn more at [rabbitholes.ai](https://rabbitholes.ai) and discover how infinite canvas thinking can supercharge your team's creative potential. ## AI Memory vs Human Memory: Why You're Still the Best Context Engine (https://www.slashspace.ai/blog/ai-memory-vs-human-memory) Published: 2026-03-24 AI memory features like ChatGPT memory and Mem0 are useful—but they're still guessing. Learn when to rely on AI memory and when human memory delivers better results. ## Introduction: The Problem With AI Memory If you've used ChatGPT, Claude, or any modern AI assistant, you've probably noticed memory features popping up everywhere. Tools like **ChatGPT Memory**, **Mem0**, and **SuperMemory** promise to remember your preferences, past conversations, and project details. But here's the truth: **all AI memory is a mathematical guess.** No matter how sophisticated the algorithm, AI memory systems are essentially asking: *"What's the most relevant piece of text to include so the AI can answer accurately?"* Sometimes they get it right. Often, they don't. The best AI users understand a critical principle: **AI is only as good as the context you give it.** Your prompt matters, but without the *right* context your results will be subpar. ## How AI Memory Actually Works Before diving deeper, it helps to understand the basics. AI memory systems typically: 1. **Embed your conversations** into vector databases 2. **Run similarity searches** when you ask a new question 3. **Retrieve chunks** of past context that *seem* relevant 4. **Package that context** into your prompt behind the scenes This process is called **retrieval-augmented generation (RAG)**—and while it's clever engineering, it's fundamentally a guessing game. The algorithm doesn't *understand* your project the way you do. It's pattern-matching, not reasoning. ## AI Memory vs Human Memory: A Practical Comparison So when should you rely on AI memory, and when should you trust your own? Here's a breakdown: | Use Case | AI Memory | Human Memory | |---|---|---| | **Personalized context** (job, location, preferences) | ✅ Very useful | Can do, but wastes mental energy | | **Project-level information** | ✅ Useful for high-level details | ✅ Better for deep context | | **Query-specific insights** | ❌ Often misses the mark | ✅ Excellent | | **Pulling relevant files or past conversations** | ❌ Hit or miss | ✅ You know exactly where to look | ### Where AI Memory Excels AI memory shines with **persistent, rarely-changing information**: * Your name, role, and company * Your preferred writing tone * Technical constraints (e.g., "I use Python 3.11") * Timezone and location This saves you from repeating yourself in every conversation. ### Where Human Memory Wins For **query-specific context**, humans are far superior: * Recalling a specific insight from a past conversation * Knowing which file contains the relevant information * Understanding *why* a piece of context matters for this exact question * Connecting dots across different projects or timeframes **Example:** Imagine you're asking an AI to help refine a proposal. AI memory might pull generic context from past chats. But *you* remember the specific feedback your manager gave three weeks ago—and you know it's exactly what the AI needs to give you a useful answer. ## The Best Approach: Human-First, AI-Assisted The most effective AI interactions follow a **human-first model**: 1. **Human memory acts as primary intelligence** – You decide what context matters 2. **AI memory serves as a fallback** – It handles the "low-hanging fruit" (persistent preferences, basic details) This isn't about rejecting AI memory features. It's about understanding their limitations and staying in control. ## What to Look for in an AI Chat App The best AI tools empower you to **control context at every step**. Look for apps that let you: * ✅ Attach files and images to your chat * ✅ Reference previous conversations as context * ✅ Select *which* past chats or documents the AI sees * ✅ Control what gets included at each point in the conversation This is called **context engineering**—and it's the difference between mediocre AI outputs and exceptional ones. Optimize your *token-to-desired-output ratio*. Every token sent to the AI should earn its place. ## Key Takeaways * **AI memory is a guess**, not a guarantee. It retrieves what *seems* relevant, not necessarily what *is* relevant. * **Human memory excels at query-specific context.** You know your projects, your goals, and your history better than any algorithm. * **Use AI memory for persistent information** like preferences, roles, and constraints. * **Stay in control.** The best AI apps let you decide what context gets included—don't outsource that decision entirely to an algorithm. * **Context engineering is a skill.** The better you get at feeding the right information to AI, the better your results. ## Finding the Right Tools Apps that give you this level of control are still surprisingly rare. Most AI interfaces either automate memory entirely (taking the decision out of your hands) or offer no memory at all (forcing you to repeat yourself constantly). We built Rabbitholes with this philosophy in mind—human memory as the primary driver, with features that let you pull in exactly the context you need, when you need it. Whether that's a previous conversation, a file, or a specific chunk of text, you stay in control of what the AI sees. It's not about having the fanciest memory system. It's about having the right context at the right moment—and often, you're the best judge of that. ## Conclusion: You're Still the Best Context Engine AI memory features are useful tools—but they're not magic. They're mathematical approximations that sometimes miss the mark. The most effective AI users treat these features as **assistants, not replacements**, for their own judgment. Your mind is still remarkably good at pulling the right context from the past, connecting it to the present question, and knowing what matters. **The best AI interaction is biased toward human intelligence.** Let AI handle the routine details. Keep the strategic context decisions for yourself. ## Why AI Writes Better Code Than Copy: Your Work Needs a Repository (https://www.slashspace.ai/blog/ai-needs-a-repository-for-knowledge-work) Published: 2026-10-11 Software engineers get sharper AI output because their AI works inside a codebase full of examples. Here is how that works under the hood, and how any kind of knowledge work can get the same compounding effect. Ask an AI coding tool to "add a settings page" inside a real software project and something quietly impressive happens. Before it writes a line, it goes looking. It opens the existing pages, finds how buttons are styled, notices which naming conventions the team uses, reads the file that explains the project's rules, and only then writes code that looks like it belongs. Ask a chat app to "write the next scene where Mara confronts her brother" and you get something fluent, generic and wrong. Mara sounds like every other fictional character. The brother has the wrong name. The prose has none of the rhythm of your first eleven chapters, because the AI has never seen them. Same kind of model. Wildly different results. The difference is not intelligence. It is that the programmer's AI is working inside a **repository**, and yours is working in a blank room. This article explains, from the ground up, why that matters: what is happening in the model, in the software wrapped around it, and in the text it actually reads. Once you see the mechanism, you can apply it to any kind of work, whatever tool you use. ## Start with the human brain Think about how an experienced professional actually produces good work. A senior copywriter does not write a product launch email from pure talent. They remember the last three launches, which subject lines worked, how the brand talks, the phrases the founder hates. A novelist on book three keeps a series bible: character names, eye colors, the way one character never uses contractions. A researcher returning to a project rereads their own notes before writing a single new sentence. Cognitive science has a useful frame for this. Your **working memory**, the part of your mind holding what you are thinking about right now, is small. Research going back decades suggests it holds only a handful of chunks at once. Experts are not experts because their working memory is bigger. They are experts because they have a huge store of **long-term memory** organized so that the right piece surfaces at the right moment. The classic studies of chess masters showed exactly this: masters could recall real game positions at a glance, but did no better than novices on randomly scattered pieces. Their advantage was a library of patterns, not raw capacity. So good human work is roughly: 1. A small, focused working memory, 2. fed by a large personal archive, 3. with a good habit of pulling the *relevant* parts of that archive in before starting. Hold on to that shape. AI follows the same one, almost exactly. ## What the model knows, and what it does not A large language model (LLM) like Claude, GPT or Gemini is trained on an enormous amount of text. During training, patterns from that text are compressed into billions of numbers called **weights**. The weights are the model's long-term memory: grammar, facts, styles, how arguments are structured, how a Python function usually looks. Two properties of the weights matter for everything that follows. **They are general.** The model has read millions of marketing emails, so it knows what a marketing email is in the average sense. It has never read *your* marketing emails. It knows how fantasy novels tend to go. It does not know that in your novel, magic costs memories. **They are frozen.** Nothing you type into a conversation changes the weights. The model does not "learn" your book by chatting with you, the way a colleague would. When the conversation ends, nothing carries over unless some software deliberately carries it over. So where does anything specific to you come from? Only one place: the **context window**. ## The context window is the model's working memory Every time you send a message, the model receives a long stretch of text, measured in tokens (roughly, pieces of words). That text includes hidden instructions from the app, the conversation so far, any files or documents the app attached, and your new message. This bundle is the context. The model reads all of it, then predicts its reply one token at a time. The context window is the model's working memory. It is much larger than a human's (modern models accept hundreds of thousands of tokens, the length of several novels), but it has the same essential role: it is the only place where *your* specifics exist for the model. Here is the part most people miss. Models are remarkably good at **in-context learning**. Show a model a few examples of a pattern inside the context, and it will continue that pattern, with no retraining at all. This was one of the headline findings when GPT-3 was introduced in 2020, in a paper literally titled "Language Models are Few-Shot Learners." Give the model three examples of your product descriptions, and the fourth comes out in your voice. Give it two scenes of Mara's dialogue, and her third scene sounds like her. Mechanically, this works through **attention**. As the model generates each new token, it looks back across the whole context and weighs which earlier tokens are most relevant to what comes next. If your context contains Mara saying "I'm not angry. I'm keeping score," then when the model writes Mara's next line, the attention mechanism can lean heavily on that line's rhythm, vocabulary and attitude. Examples in context act like a gravitational pull on the output. This is why **examples beat instructions**. You can write "use a dry, understated tone" and the model will produce its average idea of dry and understated. Show it 800 words that *are* your dry, understated tone, and it produces something far closer, because it is pattern-matching against the real thing rather than interpreting an adjective. The pantograph below makes the same point with a machine. Its linkage never changes, the way the model's weights never change. All that changes is what the stylus is given to trace. But the context also has limits worth knowing: - **It is finite.** Even a very large window cannot hold everything you have ever made, and costs and response time grow with size. - **Position matters.** Research on long contexts (for example the 2023 study "Lost in the Middle") found that models use information at the beginning and end of a long context more reliably than material buried in the middle. - **Noise dilutes signal.** If the context is stuffed with loosely related material, attention spreads thin, and the useful examples pull less hard. More context is not automatically better context. So the quality of the output depends heavily on a question the model cannot answer by itself: *what goes into the context?* ## The harness decides what the model sees The model on its own is a function: text in, text out. Everything else, including which text goes in, is decided by the software around it. Engineers call that software the **harness**. A plain chat app is a minimal harness. It sends a system prompt, your conversation history and whatever you pasted. That is all. A modern coding agent is a rich harness. When you give it a task, it runs a loop: 1. **Read the task.** 2. **Search.** It uses tools to list folders, search for keywords and open files across the project. 3. **Read what it found.** Relevant files are pulled into the context. 4. **Read the rules.** Many projects contain a file of conventions (often named something like `AGENTS.md` or `CLAUDE.md`) that the harness loads automatically: "we name things this way, never do that, test with this command." 5. **Act,** writing code that now has dozens of real examples sitting in its working memory. 6. **Check and repeat,** running the tests, reading the errors, adjusting. Notice what the harness is doing. It is playing the role of the expert's habit from earlier: *before starting, pull the relevant parts of the archive into working memory.* The model supplies the intelligence. The harness supplies the memory retrieval. The codebase supplies the archive. That combination is why AI coding has improved so dramatically in practice. The models got better, but a large share of the leap came from harnesses that let models go and read the project before they write. Here is that loop as a bench instrument. The drawer is the project, the gripper is the harness, and the small holder is the context window. Leave it alone and it fetches what is relevant on its own; click a card to hand it a specific reference. ## Why the codebase is such a good archive Software projects happen to be almost perfectly shaped for this. - **Everything is in one place.** A repository holds every file of the project in one folder tree. - **It is plain text.** The model can read all of it directly. - **It is full of precedent.** Need a new page? There are twenty existing pages to imitate. Need to handle an error? The project already handles errors in a particular way, dozens of times. - **It is searchable.** Names are consistent, so a search for "invoice" finds every relevant piece. - **It compounds.** Every feature that gets added becomes an example for the next one. Increasingly, a lot of that code was written by AI and then corrected by humans, so the archive steadily encodes the team's corrections. That last point is the quiet superpower. A codebase is a record of every decision the team made about how things should be done. Each new AI task benefits from all of them, for free, because the harness can go and look. ## Knowledge work has no repository Now look at how most writing, marketing, research and strategy work is stored. The brand voice lives in a PDF from two years ago. The best-performing emails are buried in an email tool. Chapter drafts are spread across three documents, one of them called "FINAL final v2". Research notes are in a notes app, sources in a browser bookmark folder, and the clever thing you figured out last Tuesday is in a chat thread you will never find again. And when you open a chat app to get help, the AI starts from zero. It has its frozen, general knowledge and whatever you remember to paste in. Some apps add a "memory" feature, but that is usually a short list of facts about you, like "prefers British spelling", not your actual body of work. A list of facts about a writer is not the same as reading the writer. So non-engineers end up doing the harness's job by hand: hunting for old examples, copying them in, re-explaining the project, every single time. Or, more often, they skip it, and conclude that AI output is generic. It is generic. It has nothing specific to be specific about. ## What a repository for knowledge work looks like The fix is not a smarter model. It is giving the model the same thing programmers give it: a place where all the work for a project lives together, that the AI can read from. Concretely, a knowledge-work repository would hold: - **Finished work.** Published posts, final chapters, shipped campaigns, approved reports. - **Work in progress.** Drafts, outlines, half-formed ideas. - **Source material.** Interviews, articles, transcripts, data, competitor pages. - **Rules and taste.** A style guide, a character bible, a "words we never use" list. The equivalent of a codebase's conventions file. - **The conversations themselves.** Past AI sessions where you argued with the model, rejected drafts, and explained why. Those corrections are some of the most valuable material of all. Then picture the same workflow as the coding agent: **The novelist.** You have eleven chapters, a character bible and a few scenes you wrote by hand to nail Mara's voice. You ask for the confrontation scene. Instead of writing from its average idea of a confrontation, the AI reads the character bible (so the brother's name is right), reads the scenes where Mara speaks (so her dialogue has her clipped, wry rhythm), reads the last chapter (so the scene picks up where things left off) and then writes. You still edit it. But you are editing something that sounds like your book, instead of rewriting something that sounds like the internet. **The marketer.** Your project holds the last twelve launch emails, their open rates, the brand voice doc and the new product's spec. You ask for a launch sequence. The AI can see which subject-line patterns worked, how your brand handles pricing language, and the feature names exactly as the product team spells them. **The researcher.** Your project holds forty papers, your annotations, and two earlier literature reviews you wrote. You ask for a summary of where the field disagrees. The AI works from your sources and your framing, and can cite the actual documents in front of it rather than half-remembered general knowledge. In every case, the model is identical. What changed is the archive and the harness's ability to pull from it. ## Two ways to feed context: ambient and pinpoint There are two distinct ways a project's past work can shape the output, and good knowledge work needs both. **Ambient context** is the AI being aware of everything in the project and deciding for itself what to read. This is what a coding agent does when it searches the repository. You do not have to specify anything; the project as a whole shapes the output. It is great for consistency: names, facts, general voice. **Pinpoint context** is you choosing exactly which pieces the AI reads for this particular task. Programmers do this too, by pointing the agent at a specific file: "do it like the billing page." For knowledge work, this is where the real craft lives: - "Write this scene, and use *this* earlier scene as the model for how they argue." - "Draft the announcement in the voice of *this* post, not the others." - "Summarize the debate using only *these* three papers." Pinpoint context matters because ambient context averages. If your project holds both your punchy social posts and your formal investor updates, an AI reading everything may blend them into something in between. Pointing at the right two examples gives the attention mechanism a strong, clean signal instead of a blurry one. It also keeps the context small, which, as we saw, helps the model use it well. A good setup lets you move between the two freely: let the AI read the room by default, and reach in to hand it the exact reference when precision matters. ## The compounding effect Here is why this is more than a convenience. In a chat app, using AI more does not make it better at your work. Session one hundred starts from the same blank room as session one. In a repository, every piece of work that lands in the project becomes material for the next one. The scene you rewrote by hand becomes the voice reference for the next scene. The campaign that performed well becomes the template the next one learns from. The draft where you told the AI "no, never open with a question" becomes a correction it can read next time. And it does not matter whether a given piece was written by you or by the AI. What matters is that it was *accepted*: edited, approved, kept. The project becomes a growing record of your taste, built partly from your own writing and partly from AI work you shaped into your own. The more you use it, the more the output sounds like you, because there is more of you in the archive. This is the same loop that makes AI coding feel like it is getting smarter inside a mature codebase. It is not the model getting smarter. It is the archive getting richer. The last figure runs the loop forward. Every accepted draft goes back into the drawer, so each session starts with one more example of your work than the one before. ## Where this goes wrong A repository is not magic, and it has failure modes worth knowing. - **Garbage in, imitated out.** The AI copies what it sees. If the project is full of weak drafts you never cleaned up, it will happily imitate those. Delete or separate what you would not want copied. - **Stale material.** If the old pricing page is still in the project, the AI may quote old prices. Keep a clear line between current and archived. - **Too much at once.** Stuffing everything into the context dilutes the signal. This is exactly when pinpoint context earns its keep. - **Rules nobody wrote down.** Your taste only reaches the AI if it is in the project somewhere: as examples, as a style note, or as a recorded correction. Taste that lives only in your head stays there. - **Privacy.** Whatever the AI reads may be sent to the model provider. Know what is in your project before you point a cloud model at it. ## How to start, in any tool Even without special software, you can capture most of the benefit: 1. **Pick one project** and gather its best finished work into a single place. 2. **Write a one-page rules file**: voice, banned words, names, recurring facts. Treat it like a codebase's conventions file. 3. **Before every request, attach the two or three most relevant examples,** not everything. 4. **Save accepted outputs back into the project,** along with any correction you had to make. 5. **Prune regularly.** Remove what you would not want imitated. You will notice the difference within a week. You will also notice how much manual shuffling it takes, which is the problem a harness exists to solve. ## Why Slashspace is built this way Slashspace was designed around this exact idea: each canvas is a project's repository, and the AI works inside it. Your drafts, notes, documents, web pages, videos, images and past conversations sit side by side as nodes, and the canvas is saved as a plain folder on your computer that you own. By default, a chat on a canvas gets a summary of everything there, and in Agent mode it can open any node and read it in full: that is the ambient context. When you want precision, you draw a connection from a specific scene, email or paper to the chat, or `@`-mention it, and the chat reads exactly that: the pinpoint context. Groups let you hand over a whole set of references at once, and context flows through chains, so a branch inherits everything its parent conversation could see. Every chat the AI has, and every edit you make, stays on the canvas as material for the next one. You can run it on the AI you already have, whether that is your Claude or ChatGPT plan, your own API keys or local models. Programmers stopped prompting into a void years ago. Their AI reads the project first. There is no reason your writing, marketing or research should get less: give the AI a place to read your work, and it starts sounding like you. ## The Best AI Canvas Apps in 2026, Compared (https://www.slashspace.ai/blog/best-ai-canvas-apps-2026) Published: 2026-10-24 The best AI canvas apps in 2026 compared: Slashspace, Flowith, Heptabase, Storyflow, Miro, Poppy AI, Whimsical and FigJam, on AI context, models and price. An AI canvas app puts your notes, sources and AI chats on one infinite board, so you can see your whole problem at once instead of scrolling up a single chat thread. The best ones do more than add a chatbot to a whiteboard. They let the AI read what is on the board, and they let you decide what it reads. That last part is where these apps differ most. Some AI reads the whole board, some reads only what you select, and some reads only what you connect to it. Some run one built-in model, while others let you bring your own keys, your own subscription or a model on your own computer. This guide compares eight AI canvas apps on exactly those points, with current prices, and says plainly who each one is for. **Disclosure:** we make Slashspace, one of the eight apps. Praneeth, Slashspace's founder, has used six of the eight apps himself. We've described Slashspace with the same limits we list for everyone else, and the order of the list below is not a ranking. **Quick picks** - **Deep solo research with control over context and models:** Slashspace - **A hosted canvas with AI media:** Flowith - **A long-term knowledge base on every device:** Heptabase - **Creative planning:** Storyflow - **Team workshops:** Miro - **Turning research into content and ads:** Poppy AI - **Diagrams and mind maps on a free plan:** Whimsical or FigJam ## How did we choose and judge these AI canvas apps? Praneeth, Slashspace's founder, has used six of the eight apps himself: Slashspace, Flowith, Heptabase, Miro, Whimsical and FigJam. He hasn't used Poppy AI or Storyflow, so those two are described from their own documentation and pricing pages. For all eight, the features and prices below come from each company's own homepage, feature pages, docs and pricing pages, which we read on October 24, 2026. Prices change often, and every one links to its source at the end. To make the list, an app had to give you a free-form visual canvas with AI built in, and still be open to new users. We judged each one on six questions: 1. **What can the AI see?** The whole board, a selection, or only what you connect to it. 2. **Can you control the AI's context?** Being able to choose what each answer reads matters more on a big board than on a small one. 3. **Which models can you use?** One built-in model, a menu of hosted models, your own API keys, or local models. 4. **Where does your work live?** In the company's cloud, or in files on your own computer. 5. **Which platforms?** Browser, desktop, mobile. 6. **What does it cost,** and is there a free plan? There are two kinds of "AI canvas" in search results right now. One is the generative media canvas, where nodes chain image and video models together. The other is the thinking canvas, where you lay out ideas, sources and conversations. This guide is about the second kind. If you want the first, see the note near the end. ## What are the best AI canvas apps in 2026? Here is the short version, grouped by what each app is built for. - Slashspace logo **Slashspace:** a desktop AI canvas for deep research and learning, where you connect sources to chats and branch any answer. - Flowith logo **Flowith:** a hosted AI canvas with many models plus image, video and agent tools. - Heptabase logo **Heptabase:** a visual knowledge base of cards and whiteboards, with an AI agent and apps on every platform. - Storyflow logo **Storyflow:** a browser canvas for creative projects and planning, where the AI reads your active board. - Miro logo **Miro:** the team whiteboard, now with AI that can use the board as context. - Poppy AI logo **Poppy AI:** a board for content marketers that turns research into scripts and ads. - Whimsical logo **Whimsical:** diagrams, flowcharts and mind maps, generated with AI. - Figma logo **FigJam:** Figma's whiteboard, with AI for templates, sorting and summaries. ## How do these AI canvas apps compare?
App What the AI reads Models Where your work lives Platforms Price
Slashspace logoSlashspace What you connect to each chat Hosted models, your own API keys, Claude or ChatGPT subscription, local models via Ollama Folders on your computer Mac, Windows, Linux $89/year (3-day card-required trial) or $249 one-time lifetime
Flowith logoFlowith The node you follow up from, or nodes you quote 40+ hosted models Cloud Browser Free Starter; Pro $19.90/month
Heptabase logoHeptabase Your sources, notes and cards Gemini on Pro; GPT, Claude and Gemini on Premium Cloud sync, with offline access Mac, Windows, Linux, iOS, Android Pro $8.99/month billed yearly; 7-day trial
Storyflow logoStoryflow Your active board, plus attached documents and boards Built in Cloud Browser Plus $7.99/month billed yearly; free plan announced for late 2026
Miro logoMiro The canvas, as Miro describes it Built in, on AI credits Cloud Browser (see site for apps) Free (3 editable boards); Starter $8/member/month billed yearly
Poppy AI logoPoppy AI The videos, ads, PDFs and links on your board Poppy's own model lineup Cloud Browser $1 for 7 days, then Base $399/year
Whimsical logoWhimsical Your prompt, and questions about your workspace Built in, on AI credits Cloud Browser, Mac and Windows apps Free; Pro $10/editor/month billed yearly
Figma logoFigJam Your prompt and the stickies you select Built in, on AI credits Cloud Browser (see site for apps) Free Starter; Collab seat from $3/month
"What the AI reads" is each company's own description, not our test. Prices are the lowest paid tier on each pricing page on October 24, 2026; see each section for details. ## Which AI canvas app is best for you? - **Researchers:** Slashspace if you want to control exactly which sources each answer reads and keep your work as files on your own computer. Heptabase if you want a long-term knowledge base with PDF annotation and apps on your phone. - **Students:** Heptabase for building a study base you keep for years. Whimsical or FigJam if you mostly need diagrams and mind maps and want a free plan. - **Polymaths and curious generalists:** Slashspace, because branching lets you chase three tangents from one answer without losing the main thread. Flowith if you also want image and video generation in the same place. - **Business owners and teams:** Miro for real-time team workshops. Storyflow for planning projects and campaigns on a shared board. Poppy AI if your main job is turning research into content and ads. ## Slashspace {/* FOUNDER NOTE: pending Praneeth's notes */} Slashspace is our app, a desktop AI canvas for long, exploratory work like research, learning and writing. Every chat, note and source is a node on an [infinite canvas](/features/infinite-canvas).
A Slashspace research canvas: a TED talk, a web article and a research paper wired into a chat, with two branch chats off its answer
Image: Slashspace
**What its AI sees:** only what you connect. You drag a line from a PDF, a web page or a note into a chat, and that chat reads it. Context follows chains, so a synthesis chat can read three sources while the side question next to it reads none of them. The [connections docs](/docs/canvas/connections) explain how context flows. **What stands out:** - **Branching:** fork any chat into a new connected chat, or select one sentence in an answer and ask about it. Branches keep reading their parent, so you can follow a tangent and come back. See [branching](/docs/canvas/branching). - **Model choice per node:** every chat node keeps its own model, so one canvas can put two models side by side on the same prompt. Models can come from Slashspace's hosted options, your own API keys, local models through [Ollama](/docs/settings/ollama), or a subscription you already pay for. Slashspace drives the official Claude Code and Codex CLIs on your computer, signed in with your own account. Details are in [choosing a model](/docs/ai-features/models). - **Your files:** every canvas is a plain folder on your disk that you own and can back up ([canvases and folders](/docs/canvas/canvases-and-folders)). - **Agent mode and connectors:** chats can use tools, sub-agents and MCP servers when a task needs more than one answer. **Price:** $89/year (3-day card-required trial) or $249 one-time lifetime. AI usage through your own keys, subscription or local models is paid to those providers. **Best for:** researchers, students and polymaths who want to see the whole problem at once and decide what each answer reads. **Watch out for:** it's a desktop app for Mac, Windows and Linux, with no web or mobile version. Quick Window, which lets you ask from any app, is Mac only for now. There's no free plan, and it's built for one person's deep work, not live team whiteboarding. ## Flowith {/* FOUNDER NOTE: pending Praneeth's notes */} Flowith is a hosted AI canvas. Every prompt and response becomes a node you can continue, branch from or reference later. You can follow up from one node, or quote several nodes in a single prompt to combine outputs from different branches.
A Flowith canvas of prompt and response nodes, with generated images and a grouped set of notes
Image: Flowith blog
**What stands out:** a long list of hosted models (its pricing page says 40+), plus AI image, video, slides, website and deep research tools, and an agent called Neo. It all runs in the browser with nothing to install. **Price:** a free Starter plan with a one-time credit allowance and limited models. Pro is $19.90 a month, or $214.92 a year billed yearly. Ultimate and Infinite tiers sit above it. **Best for:** people who want one hosted app for AI chat on a canvas and AI media. Our [Flowith comparison](/compare/flowith-alternative) goes into more detail. **Watch out for:** plans run on credits, and much of the pricing page is about image and video generations, so heavy media use costs more. Your canvases live in Flowith's cloud. We didn't find bring-your-own-key or local model options documented. ## Heptabase {/* FOUNDER NOTE: pending Praneeth's notes */} Heptabase is a visual knowledge base for students, researchers and lifelong learners. Notes live as cards you arrange on whiteboards, and the same card can appear on many boards.
A Heptabase whiteboard of connected note cards about a book, with one card open in a side panel
Image: Heptabase website
**What stands out:** PDF annotation, YouTube transcripts, a web clipper, Zotero and Readwise integrations, daily journals and bi-directional links. Its AI can explain sources, take notes and organize your knowledge base, and there's an AI tutor on higher plans. Heptabase also lists a CLI that works with Claude Code and Codex. It has offline access, free real-time collaboration for people you invite, and apps for Mac, Windows, Linux, iOS and Android. **Price:** Pro is $8.99 a month billed yearly, Premium $17.99 and Premium+ $53.99, each billed yearly. **Best for:** students and researchers who want a knowledge base they curate for years, on every device. See our [Heptabase comparison](/compare/heptabase-alternative). **Watch out for:** the AI depends on your plan. Pro's AI agent uses Gemini models only, with 100 AI credits a month. Premium adds GPT and Claude. There's no permanent free plan, just a 7-day trial. ## Storyflow Storyflow is a browser-based AI canvas for planning, storyboards, moodboards and creative projects. You describe what you're working on, and its AI lays out cards, notes and groups across the board.
A Storyflow board in dark mode, laid out as a mind map of connected cards
Image: Storyflow website
**What its AI sees:** Storyflow says its AI reads the board you have open, and you can attach up to five documents and three boards as extra context. It also has an MCP server, so Claude, ChatGPT, Gemini and Grok can build and read Storyflow boards from inside those assistants. **What stands out:** a library of 200+ creative templates it calls Tactics (Hero's Journey, AIDA, StoryBrand and more), AI image generation on Pro, free collaboration for anyone a paid member invites, and image or PDF export. **Price:** Plus is $7.99 a month billed yearly, Pro $14 and Max $39, each billed yearly. **Best for:** creators, founders and small teams planning projects, campaigns or stories on a shared board. **Watch out for:** Storyflow is in paid early access, so there's no free plan yet; it says one is coming before the end of 2026. It's cloud-only, and its own review notes that there's no local or offline option. We didn't find a way to pick models or bring your own keys inside the app. ## Miro {/* FOUNDER NOTE: pending Praneeth's notes */} Miro is the best-known collaborative whiteboard, and AI is now central to how it presents itself. Miro says you can use your entire canvas as a prompt, and its Sidekicks build frames, timelines, kanbans and diagrams on the board.
A Miro board with frames, sticky notes and a document, collaborator cursors, a Sidekick and an AI reply at the bottom
Image: Miro website
**What stands out:** real-time collaboration for large teams, 7000+ templates, docs, tables and slides on the canvas, AI workflows and agents on higher plans, and MCP connections to tools like Claude, Cursor and Copilot. **Price:** Free, then Starter at $8 per member a month and Business at $20 per member a month, billed yearly. **Best for:** teams running workshops, planning and retros together. **Watch out for:** AI runs on credits per member per month: 10 on Free, 25 on Starter and 50 on Business. The free plan keeps only three boards editable. It's built for teams, which makes it a lot of tool for solo research. ## Poppy AI Poppy AI is a visual board built for creators, founders and marketers. You bring in videos, ads, web pages, PDFs, transcripts and competitor content, and Poppy helps you find hooks and patterns and turn them into scripts, ads, emails and landing pages in your voice.
A Poppy AI board with YouTube and Instagram videos connected to several AI chat panels
Image: Poppy AI website
**What stands out:** it's focused on one job, content, and does it end to end, from research on the board to finished copy and images. It remembers your audience, offer and past content in one workspace. **Price:** $1 for 7 days of full access, then Base at $33 a month billed yearly ($399 a year). Creator is $189 a month billed yearly. Poppy offers a 30-day money-back guarantee. **Best for:** creators and marketing teams whose canvas work ends in content. See our [Poppy AI comparison](/compare/poppy-ai-alternative). **Watch out for:** it's a content tool more than a general thinking canvas. Plans are credit-based, and the team plans are a big step up in price. ## Whimsical {/* FOUNDER NOTE: pending Praneeth's notes */} Whimsical is a canvas for diagrams, flowcharts, mind maps and wireframes. Its AI generates those from a prompt, and Whimsical Agent can answer questions about your workspace. A Whimsical MCP server lets Claude, ChatGPT, Cursor and Codex work with it too.
Whimsical Agent drawing a user onboarding flowchart on the canvas, with its chat panel on the right
Image: Whimsical website
**What stands out:** clean, fast diagrams, AI mind maps that branch an idea out for you, and desktop apps for Mac and Windows. **Price:** Free, then Pro at $10 per editor a month and Business at $20, billed yearly. **Best for:** students and teams who think in flowcharts and mind maps. **Watch out for:** it's built for structure and diagrams, not long conversations with sources. The free plan caps you at 50 board objects a month and 10 AI credits per member. ## FigJam {/* FOUNDER NOTE: pending Praneeth's notes */} FigJam is Figma's online whiteboard. FigJam AI generates templates for meetings, brainstorms and timelines from a prompt, sorts stickies into groups and summarizes them, and Jambot brings ChatGPT into the board for brainstorming.
The FigJam AI Generate panel suggesting templates next to a Weekly Sync board
Image: Figma website
**What stands out:** it lives inside the Figma ecosystem, so design teams can move from a FigJam session straight into their files. **Price:** Figma's free Starter plan includes FigJam with a daily AI credit allowance. On Professional, a Collab seat that includes FigJam starts at $3 a month. **Best for:** product and design teams already working in Figma. **Watch out for:** the AI helps you run and tidy a session more than it reasons over your sources. AI runs on credits that depend on your seat. ## Which apps almost made the list? - **Ponder:** an AI research workspace that turns papers, slides and recordings into mind maps and notes. It has a free plan with 50 credits a day, and paid plans from $14 a month billed yearly. It now presents itself as a research workspace more than a canvas. - **tldraw:** a source-available canvas SDK for developers, with a free whiteboard at tldraw.com. In October 2026 it launched a plugin that puts a live tldraw canvas next to ChatGPT and Codex. Its earlier Make Real project was archived in February 2026. - **Obsidian Canvas:** a free, local-first canvas inside Obsidian. Obsidian is free, including for work, but Canvas has no built-in AI. - **Excalidraw:** a free, open-source sketching canvas. The free app includes 10 generative AI requests a day, and Excalidraw+ is $6 per user a month. - **Kosmik:** a visual canvas with a built-in browser that often appears on these lists. It announced it would wind down on May 31, 2026, and no longer accepts new sign-ups. ## Is ChatGPT Canvas an AI canvas app? Not in the sense this guide uses. ChatGPT's canvas, like Gemini's Canvas, is an editor that opens next to the chat so you can work on one document or piece of code with the AI. It doesn't give you a free-form board where many chats, notes and sources sit side by side. If that's what you want, our [ChatGPT canvas comparison](/compare/chatgpt-canvas-alternative) explains the difference. ## What about AI canvases for images and video? A separate group of apps uses "AI canvas" for node-based media generation, where you chain image and video models together on a board. Tools like Figma Weave, Krea, Flora and ComfyUI belong there. Flowith straddles both groups. If your goal is visuals rather than thinking, start with those. ## Frequently asked questions ### What is an AI canvas app? An AI canvas app combines an infinite visual board with AI. You place notes, sources and AI chats on the board, and the AI can read some or all of it. The difference from a chat app is that you see your whole problem at once, instead of scrolling one long thread. ### What is the best AI canvas app in 2026? It depends on the job. For deep solo research with control over context and models, Slashspace. For a hosted canvas with AI media, Flowith. For a long-term knowledge base on every device, Heptabase. For team workshops, Miro. For creative planning, Storyflow. ### Is there a free AI canvas app? Yes. Flowith, Miro, Whimsical and FigJam have free plans with limited AI, and Ponder has a free plan with daily credits. Slashspace and Heptabase have no free plan. Storyflow says a free plan is coming before the end of 2026. ### Which AI canvas app lets me use my own API keys or local models? Of the eight here, Slashspace documents both: your own API keys from providers like OpenAI, Anthropic and Google, and local models through Ollama. It can also use a Claude or ChatGPT subscription you already pay for. We didn't find either option documented for the other seven. ### Which AI canvas app keeps my work on my computer? Slashspace saves every canvas as a folder of files on your computer. Obsidian Canvas also stores files locally but has no built-in AI. Heptabase offers offline access, and the rest are cloud apps. ### What's the difference between an AI canvas and an AI whiteboard? They overlap. "AI whiteboard" usually means a team tool like Miro or FigJam, built for live sessions with stickies and diagrams. "AI canvas" more often means a board where AI chats and sources live together and the AI reads what's there, like Slashspace, Flowith or Storyflow. ### Can an AI canvas replace ChatGPT? For exploratory work, often yes, because you can keep many threads, sources and branches visible at once. For a quick question, a regular chat is still faster. Many people use both. If you branch a lot, our guide to the [best branching AI chat apps](/blog/best-branching-ai-chat-apps-2026) covers the chat side. ## The bottom line The right AI canvas app depends on what you want the AI to see. If you want it to see the whole board and your team with it, Miro and Storyflow are built for that. If you want a knowledge base you curate for years, Heptabase. If you want AI media alongside your canvas, Flowith. If you want to see your whole problem at once and decide exactly what each answer reads, with your choice of models and your files on your own computer, that's what we built Slashspace for. You can [download it](/download) or compare [plans](/pricing). ## Sources Every page below was read on October 24, 2026. - Slashspace: [why a canvas](/docs/why-a-canvas), [branching](/docs/canvas/branching), [connections](/docs/canvas/connections), [models](/docs/ai-features/models), [canvases and folders](/docs/canvas/canvases-and-folders), [pricing](/pricing) - Flowith: [pricing](https://flowith.io/pricing) - Heptabase: [pricing](https://heptabase.com/pricing), [homepage](https://heptabase.com) - Storyflow: [pricing](https://storyflow.so/pricing), [infinite canvas](https://storyflow.so/infinite-canvas), [MCP](https://storyflow.so/mcp), [its infinite canvas tools review](https://storyflow.so/blog/best-infinite-canvas-tools-2026) - Miro: [pricing](https://miro.com/pricing/), [Miro AI](https://miro.com/ai/) - Poppy AI: [homepage and pricing](https://getpoppy.ai/) - Whimsical: [pricing](https://whimsical.com/pricing), [Whimsical AI](https://whimsical.com/ai) - FigJam: [Figma pricing](https://www.figma.com/pricing/), [FigJam AI](https://www.figma.com/figjam/ai/) - Ponder: [pricing](https://ponder.ing/pricing) - tldraw: [ChatGPT plugin post](https://tldraw.dev/blog/what-we-ve-built-with-the-tldraw-chatgpt-plugin), [Make Real repository](https://github.com/tldraw/make-real) - Obsidian: [pricing](https://obsidian.md/pricing) - Excalidraw: [Excalidraw+ pricing](https://plus.excalidraw.com/pricing) - Kosmik: [homepage](https://www.kosmik.app) (wind-down notice) ## Best AI Mind Mapping Tools in 2026 (Ranked and Compared) (https://www.slashspace.ai/blog/best-ai-mind-mapping-tools-2025) Published: 2026-10-21 The best AI mind mapping tools in 2026 compared: Slashspace, Heptabase, MyMap, Flowith, Miro, Whimsical and Taskade. What each does well, where it falls short, and who it suits, from each tool's own pages. The best AI mind mapping tools in 2026 don't just draw a diagram from a prompt. They let you think with AI on a canvas: branch an idea, pull in real sources, and keep every direction you tried in view. For research and hard problems, **Slashspace** is our pick. For fast diagrams, **MyMap** and **Whimsical** are quicker. For team workshops, **Miro** is the natural choice. "AI mind mapping" now covers very different tools. Some are classic mind mappers or whiteboards with an AI assistant. Others are visual AI workspaces, where the canvas holds conversations and sources instead of boxes and lines. That difference matters more than any feature list, so this guide covers both. ## How we put this list together We read each tool's own product, help and pricing pages on October 21, 2026, and the Heptabase and Flowith details come from our comparison pages, checked on October 2, 2026. We didn't run hands-on tests for this roundup, so features are as each company describes them, and they change often. We make Slashspace, so weigh our first place accordingly. We've tried to say plainly where other tools are the better fit. ## Quick comparison | Tool | What it is | AI on the canvas | Sources as context | Best for | |---|---|---|---|---| | **Slashspace** | Desktop AI canvas (Mac, Windows, Linux) | Chats are nodes you branch; any model, including your own keys and local models | PDFs, Office files, web pages, YouTube and social posts, images, notes | Research, learning and hard problems | | **Heptabase** | Card and whiteboard knowledge base | AI agent that edits cards and whiteboards | PDFs, Word, web cards, YouTube | Building a long-term knowledge base | | **MyMap** | Web AI canvas that draws diagrams | Chat that draws and edits mind maps, flowcharts and more | Pasted notes, PDFs, Word docs, YouTube links | Turning a topic into a diagram fast | | **Flowith** | Web AI canvas | Parallel model outputs, Agent Neo | Files, links, YouTube transcripts | Generating content, slides and media | | **Miro** | Team whiteboard | Miro AI generates and expands mind maps | Board content, files dropped on the board | Team brainstorms and workshops | | **Whimsical** | Diagramming and docs workspace | Whimsical Agent generates mind maps and flowcharts | Your workspace | Clean diagrams for product teams | | **Taskade** | Project workspace with a mind map view | AI expands branches and turns them into tasks | Projects, docs and links | Turning a brainstorm into a plan | ## What makes an AI mind mapping tool worth using? **A real canvas.** Two dimensions let proximity, grouping and layout carry meaning that a scrolling chat throws away. **AI built into the canvas.** You should be able to talk to the AI where your ideas live, not copy and paste between a chat window and a diagram. **Context you control.** Can you give the AI your own documents, links and notes, and decide what each question reads? Answers grounded in real sources are far more useful than answers from a prompt alone. **A choice of models.** Different models are good at different things. Being able to pick, and to compare two answers to the same question, gives you more control over quality and cost. **Branching without losing your place.** Exploration means trying directions. A good tool lets you follow a tangent and keep the original thread intact. ## The top AI mind mapping tools in 2026 ### 1. Slashspace **Best for: research, learning and any problem you think through with AI over days or weeks** [Slashspace](/) is a desktop AI canvas for Mac, Windows and Linux. Every chat is a node on an infinite canvas, and any chat can branch into new chats. Instead of one thread that grows downward, you get a map of your thinking, where each branch is a direction you chose to explore. It isn't a classic mind mapper that turns a prompt into a diagram. It maps the process of thinking something through with AI. **What sets it apart:** - **Branch from any message or sentence.** Send a follow-up into a new branch, or select one sentence of an answer and ask about it. The original thread stays as it was. See [branching](/features/branching). - **Sources you wire in.** PDFs, Office files, web pages, YouTube, TikTok, Instagram, X and LinkedIn posts, images and notes become nodes. Connect a source to a chat and that chat reads it in full. Unconnected chats don't. - **Any model, side by side.** Each chat node has its own model. Use Slashspace's hosted models, your own API keys, your Claude or ChatGPT subscription through Claude Code or Codex, Cursor, or local models through Ollama. Branch a question once per model and the answers sit next to each other. See [compare models](/features/compare-models). - **Orchestrator.** Give it a goal and it builds the canvas for you: question chats, notes and sources, already connected. - **New in 4.0.** Quick Window lets you ask from any app on your Mac and drops the answer onto a canvas. Stacks fold finished nodes into decks so big canvases stay calm and fast. Replays (beta) let you share how a canvas came together. See [Slashspace 4.0](/blog/slashspace-4-0). - **Local-first.** Canvases are plain folders on your computer, and you can download any canvas as Markdown or JSON. **Where it falls short:** it's a desktop app, so there's no web or phone app today, and there are no shared, multiplayer canvases yet. If you want a finished diagram to hand in, a diagram generator is faster. **Pricing:** $89/year, starting with a 3-day trial (card required), or $249 one-time lifetime. See [pricing](/pricing). ### 2. Heptabase **Best for: building a card and whiteboard knowledge base over years** Heptabase is a visual note-taking app built around cards arranged on whiteboards. Its AI has grown a lot. An AI agent can edit cards and whiteboard connections, you can choose which cards, whiteboards or paragraphs it reads, and answers link back to the exact passage or video timestamp. It also reads PDFs, Word files, web cards and YouTube, and syncs with Zotero. **Strengths:** - A mature card, tag, journal and backlink knowledge base - PDF highlighting and annotation, plus Zotero sync, for academic reading - Web, iOS and Android apps with cloud sync and shared whiteboards **Limitations:** - No chat branching and no side-by-side model comparison - Custom AI endpoints must be public HTTPS, so local models through Ollama aren't possible - Subscription plans only **Best for:** people who already keep a substantial note-taking practice and want AI to help them connect and work with what they've collected. See our [Heptabase comparison](/compare/heptabase-alternative). ### 3. MyMap **Best for: turning a topic or a document into a diagram quickly** MyMap is a web AI canvas where you talk through an idea and the AI draws it as a mind map, flowchart, SWOT, timeline or one of many other diagram types. You can bring in pasted notes, a PDF, a Word document or a YouTube link, select a node and ask for a change, and frame parts of the board as slides. Exports include PNG, SVG, PDF, Markdown and share links. **Strengths:** - Picks the diagram type for you and lays it out - Many diagram formats from one chat, including genograms and ERDs - A free tier, with premium models on paid plans **Limitations:** - Built to produce a diagram, not to keep many separate lines of questioning side by side - Works with the models MyMap offers on each plan, in the browser **Best for:** students, consultants and anyone who needs a clean map, flowchart or deck to hand in. ### 4. Flowith **Best for: generating content, slides and media on a canvas** Flowith is a web AI canvas that runs prompts across 40+ hosted models and lays parallel outputs side by side. Agent Neo plans subtasks as nodes and can build slides and webpages, and Flowith also generates images and video. Knowledge Garden holds your files and links. **Strengths:** - Strong at generating finished outputs: decks, pages, images and video - Web and iOS apps, plus share links with edit roles - A free plan to start **Limitations:** - Hosted models only: no API keys of your own and no local models - Branches whole nodes rather than single sentences **Best for:** people who want an AI canvas for producing content more than for long research. See our [Flowith comparison](/compare/flowith-alternative). ### 5. Miro AI **Best for: teams that already brainstorm in Miro** Miro is one of the best known collaborative whiteboards. Its mind map tool now includes Miro AI, which generates a multi-branch mind map from a prompt or pasted text, expands a branch when you're stuck, and offers structures for concept exploration, user needs, problem analysis and strategic planning. Maps live on the same board as everything else and can be presented or shared. **Strengths:** - Real-time and async collaboration, comments and templates - Connects to 250+ work tools - A free plan with AI features to try **Limitations:** - AI helps with the map rather than holding long conversations with your sources - Made for groups more than for solo deep work **Best for:** teams running workshops and brainstorms who want AI help in a familiar space. ### 6. Whimsical AI **Best for: fast, clean diagrams for product teams** Whimsical has long been a favorite for wireframes and flowcharts. Whimsical Agent generates flowcharts, wireframes and mind maps, or answers questions about your workspace. Whimsical MCP lets you use it from Claude, ChatGPT, Cursor, Codex and other agents. **Strengths:** - Quick, polished diagrams - Works from the AI tools your team already uses, through MCP **Limitations:** - Mind maps are one feature among many - Not built for long, source-heavy research **Best for:** product managers, designers and planners who need clear diagrams as part of their work. ### 7. Taskade AI **Best for: turning a brainstorm into a plan** In Taskade, a mind map is one view of a project. The AI generates and expands branches from a prompt, text or documents, and every branch can become a task with an owner and a due date. Switch to a list, board, calendar or Gantt view and you're looking at the same data. **Strengths:** - The map is the project plan, so nothing has to be retyped - Real-time collaboration, AI agents and automations **Limitations:** - Mind mapping is one part of a broad project tool - The AI is aimed at planning and doing more than at open-ended exploration **Best for:** small teams and solo founders who want ideas and tasks in one place. ## What is a visual AI workspace? A visual AI workspace is a canvas where AI conversations, sources and notes sit side by side as objects you can arrange and connect, instead of living in one scrolling chat. A mind map shows the structure of a topic. A visual AI workspace shows the structure of your thinking about it: which questions you asked, what each answer was based on, and where you branched. The idea grew out of a simple problem. Hard questions are never one question. In a chat app, every tangent goes into the same thread, the model reads all of it, and a week later you can't find the answer that mattered. On a canvas, each tangent gets its own place, and you decide what each one sees. ### What should a visual AI workspace do? - **Hold many conversations at once,** each with its own context, so tangents don't pollute each other. - **Let you branch** from any answer without losing the original thread. - **Treat sources as first-class objects** that you connect to the chats that need them. - **Let you pick the model** for each question and compare answers side by side. - **Stay readable as it grows,** with ways to group, fold away and search what you've done. - **Leave your work in a form you own,** so you can export it and come back to it. ### How Slashspace works as a visual AI workspace In Slashspace, the canvas is the workspace: - **Chats, notes and sources are nodes.** A chat reads what is connected to it, so a synthesis chat can read three papers while the side question next to it reads none of them. - **Branches sit side by side.** Fork a chat, ask about a single sentence, or branch the same prompt to several models and read the answers together. - **Text nodes are pages.** Since 4.0, notes are portrait pages with headings, lists, quotes, code blocks and to-dos. - **Big canvases stay usable.** Group related nodes, fold finished ones into Stacks, tidy the layout in one keystroke, and search the canvas with `⌘K`. - **Work comes to the canvas.** Quick Window (Option+/ on Mac) sends a question or a note from any app straight onto a canvas. - **You can show your process.** A Replay (beta) publishes a canvas as a link where anyone can watch it build itself, step by step, with the parts you choose to show. If you mostly need a diagram, a mind map generator is the better tool. If you're working through something hard with AI and want to see and steer the whole investigation, a visual AI workspace is what you're after. ## How to choose the right tool ### If you're doing research or working through complex ideas You need to explore in steps: branch into sub-questions, bring in sources and build understanding over several sessions. **Slashspace** is the strongest fit for that. **Heptabase** is worth a look if you already keep a large note collection and want AI to work with it. ### If you need to visualize or present a topic quickly You want clean output you can share. **MyMap** and **Whimsical** are built for speed and good-looking diagrams. ### If you're running team workshops You need real-time collaboration and templates. **Miro** is made for exactly this, and **Taskade** helps if the workshop should end in a task list. ## Frequently asked questions ### What is the best AI mind mapping tool in 2026? It depends on the job. For research and long, exploratory work with AI, Slashspace. For a diagram from a prompt or a document, MyMap or Whimsical. For team brainstorms, Miro. For turning a map into a project plan, Taskade. ### What's the difference between an AI mind map tool and an AI chat tool? An AI chat tool gives you one linear thread. A mind map tool lets you arrange ideas in space and branch in many directions. A visual AI workspace like Slashspace combines the two: conversations on a canvas, each branching from the last. ### Can I use AI mind mapping tools for academic research? Yes, but look for tools that take your sources as context. Slashspace reads PDFs and other documents you connect to a chat, and Heptabase adds PDF annotation and Zotero sync. See [Slashspace for literature reviews](/for/literature-review). ### Do I need to know how to write prompts? No. Most of these tools work with plain questions. Asking clear questions and structuring your thinking matters more than prompt technique. ### Which tools work for solo users and which for teams? For solo deep work: Slashspace or Heptabase. For collaborative sessions: Miro or Taskade. For quick visual output you'll share with a team: Whimsical or MyMap. ## The bottom line AI mind mapping has moved past "AI draws a diagram for you." The most capable tools in 2026 let you think with AI, not just produce output with it. If you want a tool that follows how exploratory thinking really works (branching, grounded in real sources, free to use any model), try Slashspace. For quick diagrams, MyMap and Whimsical remain strong. For team collaboration, Miro is hard to beat. For a long-term knowledge base, Heptabase has a loyal following for good reason. [See pricing](/pricing) · [Start a 3-day trial](/checkout/yearly) Slashspace is $89/year, starting with a 3-day trial (card required), or $249 one-time lifetime, on Mac, Windows and Linux. ## Read next - [The best branching AI chat apps in 2026](/blog/best-branching-ai-chat-apps-2026) - [10 ways AI mind mapping can supercharge brainstorming](/blog/10-ways-ai-mind-mapping-supercharge-brainstorming) - [The infinite canvas in Slashspace](/features/infinite-canvas) - [How Slashspace compares](/compare) ## The Best Branching AI Chat Apps in 2026 (https://www.slashspace.ai/blog/best-branching-ai-chat-apps-2026) Published: 2026-10-08 The best branching AI chat apps in 2026: how ChatGPT, Claude, Gemini, Flowith, TypingMind, Msty, LibreChat and Slashspace branch, from each app's own docs. The best branching AI chat apps let you split a conversation at any point, follow a tangent, and keep the original thread intact. In 2026 that's no longer a niche feature. ChatGPT, Claude and Gemini all branch now, and a handful of other apps go further, with branch maps, side-by-side comparisons or whole canvases. The catch is that "branching" means something different in each one. This roundup covers eight apps that really do branch, explains how each works, and says who each one suits. ## How we put this list together We read each app's own help pages, docs and pricing pages on October 8, 2026. We didn't run hands-on tests for this roundup, so features and prices are as each company describes them, and they change often. We make Slashspace, one of the eight apps. It's included because it's built around branching, but this isn't a ranking, and we haven't put ourselves on top. The apps are grouped by type, and each section ends with who it's best for. Apps we looked at and left out: we didn't find chat branching documented for Heptabase, Poppy AI or NotebookLM (now Gemini Notebook). LM Studio documents duplicating a whole chat, which is close but not the same thing. ## Three kinds of branching - **Branch into a new chat.** You pick a message, and the app copies the conversation up to that point into a separate chat. ChatGPT, Gemini and LibreChat work this way, and TypingMind offers it too. - **Branch inside the chat.** You edit an earlier message, and the chat keeps both versions, with arrows or a dropdown to switch between them. Claude and Msty work this way. - **Branch on a canvas.** Every branch is a node on a visual map, connected to the chat it came from. Flowith and Slashspace work this way. ## Quick comparison | App | How you branch | Where branches live | Platforms | Price | |---|---|---|---|---| | ChatGPT | Branch in new chat | Separate chats | Web | Free; Plus $20/month | | Claude | Edit an earlier message | Versions inside one chat | Web and desktop | Free; Pro $20/month | | Gemini | Branch in a new chat | Separate chats | Web, Android, iPhone | Free; paid Google AI plans | | Flowith | Follow up from any node | Nodes on a canvas | Web | Free Starter; Pro $214.92/year | | Slashspace | Branch a chat or one sentence | Connected nodes on a canvas | Mac, Windows, Linux | $89/year, starting with a 3-day trial (card required), or $249 one-time lifetime | | TypingMind | Chat Thread, or Fork chat from here | Threads inside one chat, or separate chats | Web app you can install | From $39 once, plus API costs | | Msty Studio | Edit as Branch | Labeled branches inside one chat | Mac, Windows, Linux; web on Aurum | Free; Aurum $149/year per user | | LibreChat | Fork from any message | Separate chats | Self-hosted web app | Free and open source | "Platforms" means where each company documents branching, not everywhere the app runs. ## Built into the big assistants ### ChatGPT OpenAI added branching to ChatGPT on the web in September 2025. Hover over a message, open **More actions** (⋯) and choose **Branch in new chat**. You get a new chat with everything up to that message, and the original stays as it was. Inside a project, branched chats appear next to the original. In a shared project, members can branch a teammate's chat to try a new direction without changing it. **Good to know:** each branch is a separate chat in your sidebar, so after a few branches it's hard to see which came from where. A branch doesn't see anything you add to the original later. **Best for:** people who already live in ChatGPT and branch now and then. ### Claude In Claude's chat app, you branch by editing. Go back to one of your earlier messages, click **Edit**, change it and send. Claude answers from that point, and arrows under the message let you switch back to the original version. **Good to know:** you branch from your own messages, and only one version shows at a time. Branches don't appear as separate chats. If you code in the terminal, Claude Code has a `/branch` command that copies the session so far and switches you into the copy. **Best for:** quick "what if I'd asked it differently" retries. ### Gemini Google began rolling out branching in Gemini in April 2026. Below a response, open **Menu** and choose **Branch in a new chat**. Google's help page covers the web, Android and iPhone. **Good to know:** the two chats are independent. Deleting the branch doesn't delete the original, and the other way around. **Best for:** branching on your phone. Of the apps here, Gemini is the only one whose branching instructions cover Android and iPhone. ## AI canvas apps ### Flowith Flowith is a hosted AI canvas. Every prompt and response becomes a node you can continue, branch from or reference later. In its **Follow-up First** mode, clicking a node and sending a message continues from that node with the full context of the thread above it. In **Quote First** mode, you can click several nodes to quote them in one prompt, which lets you combine outputs from different branches. A duplicate button copies a node so you can take an idea in a new direction. **Good to know:** branches start from whole nodes. Flowith also bundles image and video generation, and plans run on credits. Pro is $19.90 a month, or $214.92 a year billed yearly, and there's a free Starter plan. **Best for:** people who want a canvas and AI media tools in one hosted app. See our [Flowith comparison](/compare/flowith-alternative) for the detail. ### Slashspace Slashspace is our app, a desktop AI canvas for Mac, Windows and Linux. Every chat is a node, and every branch is a new chat connected to the one it came from. - **Branch a chat:** type your message and press ⌘⇧Enter (Ctrl+Shift+Enter on Windows and Linux) to send it to a new branch, or ⌘B for an empty one. - **Branch from one sentence:** select text in any answer and ask about it, and the branch opens with that passage quoted. - **Branches stay live:** a branch keeps seeing new messages in its parent. Sibling branches don't see each other. - **Compare models:** open one branch per model and send the same prompt to each. - **Combine branches:** connect two finished branches into a new chat and it reads both in full. It runs on your own API keys, local models through [Ollama](/docs/settings/ollama), or your Claude or ChatGPT subscription. For subscriptions, Slashspace drives the official [Claude Code and Codex command-line tools](/docs/settings/subscription-providers) on your computer, signed in with your own account, and usage counts against your plan. **Good to know:** it's a desktop app, with no web or mobile version yet. It's $89/year, starting with a 3-day trial (card required), or $249 one-time lifetime. Details are in the [branching docs](/docs/canvas/branching). **Best for:** people who branch every session and want to see, compare and merge their branches. ## Chat apps with branching ### TypingMind TypingMind is a chat app you use with your own API keys. It gives you two ways to branch. **Chat Thread** keeps branches inside the same conversation: edit or regenerate a message, or start a new thread at any point, and switch between threads with the arrow buttons. Each thread uses the history up to that point and is saved on its own. Or open the three-dot menu next to any message and choose **Fork chat from here** to get a new conversation with the context up to that point. There's also Side Chat for quick questions that stay out of the main chat. **Good to know:** it's a web app you can install like a desktop app, and your data lives in your browser unless you turn on sync. Licenses are one-time: $39, $79 or $99 for Premium (listed at $198 when we checked), and you pay your AI provider separately. Premium adds multi-model chats, where each model keeps its own thread. **Best for:** bring-your-own-key users who want a one-time price. ### Msty Studio Msty Studio is a local-first desktop app for Mac, Windows and Linux. Open the actions menu on any message, yours or the AI's, and choose **Edit as Branch**. The original path is kept, and a branch dropdown lets you switch paths. Its Branch Explorer lets you label branches and compare two of them side by side as a diff, and Split Chat runs several models side by side with the same input. **Good to know:** it runs local models (Ollama, Llama.cpp and MLX) as well as online providers. There's a free plan; Aurum is $149 per user per year and adds Msty Studio Web. **Best for:** people who want local models and serious branch comparison in one desktop app. ### LibreChat LibreChat is a free, open-source (MIT) chat app that you host yourself or on a server. Open a message's menu and pick the fork icon, and you get a new conversation, with three options for what to copy: only the visible path, the path plus related branches, or everything up to and around that message (the default). A **Start fork here** option forks from that message to the end of the chat instead. **Good to know:** you need to install and run it, and you bring your own model providers. **Best for:** teams and tinkerers who want branching on infrastructure they control. ## Which branching AI chat app should you pick? - **You branch occasionally and use one assistant:** stick with the built-in option in ChatGPT, Claude or Gemini. - **You branch on your phone:** Gemini. - **You want local models:** Msty Studio, or Slashspace with Ollama. - **You want a one-time license and your own keys:** TypingMind, or Slashspace's $249 one-time lifetime option. - **You want to self-host:** LibreChat. - **You want to see every branch on a map:** Flowith or Slashspace. Flowith if you also want hosted image and video tools; Slashspace if you want to use the AI plans you already pay for and work from your own files. ## Frequently asked questions ### What is a branching AI chat app? An app that lets you split a conversation at a chosen message and continue in a new direction without changing the original thread. Depending on the app, the branch becomes a new chat, a second version inside the same chat, or a new node on a canvas. ### Does ChatGPT have branching? Yes. On chatgpt.com, hover over a message, open More actions (⋯) and choose Branch in new chat. ### Can Claude branch conversations? Yes, by editing one of your earlier messages and sending it. Arrows under the message switch between versions. ### Which apps can show all branches at once? Of the apps here, the canvas apps (Flowith and Slashspace) show every branch on one map. Msty's mini map and Branch Explorer give you an overview inside a chat. ### Can I merge two branches? Not in ChatGPT, Claude or Gemini, short of copying and pasting. In Flowith you can quote nodes from different branches into one prompt. In Slashspace you connect the branches into a new chat, and it reads both in full. ## The bottom line If you branch now and then, the buttons already in ChatGPT, Claude and Gemini are enough. If you branch every day, pick by what matters most to you: local models, a one-time price, self-hosting or a canvas map. If that last one sounds right, Slashspace is $89/year, starting with a 3-day trial (card required), or $249 one-time lifetime. [See pricing](/pricing) · [Start a 3-day trial](/checkout/yearly) ## Read next - [Linear AI chat apps versus non-linear apps](/blog/linear-ai-chat-apps-versus-non-linear-apps) - [Non-linear AI chat: 5x faster](/blog/non-linear-ai-chat-5x-faster) - [Branching in Slashspace (docs)](/docs/canvas/branching) - [How Slashspace compares](/compare) ## ChatGPT Conversation Too Long? Why Long Chats Break and What to Do (https://www.slashspace.ai/blog/chatgpt-conversation-too-long) Published: 2026-10-08 ChatGPT conversation too long? Why long chats forget early details or hit the length limit, the fixes OpenAI supports, and how to keep big projects manageable. Is your ChatGPT conversation too long? You'll usually notice in one of two ways. Either ChatGPT starts forgetting things you settled near the start (a name, a constraint, a decision), or it stops letting you add messages at all. The message people report seeing is: *"You've reached the maximum length for this conversation, but you can keep talking by starting a new chat."* Both have the same root cause, and both have fixes. This guide explains what's going on, then goes through the fixes OpenAI supports, from quickest to most thorough. At the end we cover a different way to organize long work so a single chat never has to hold all of it. ## Why long ChatGPT chats break ### The context window is a fixed budget A model can only read a limited amount of text at once, measured in tokens (in English, a token is roughly three quarters of a word). That limit is the **context window**. Everything ChatGPT considers when it answers has to fit inside it. OpenAI publishes the total window for each plan. As of October 8, 2026, ChatGPT's pricing page lists: | Plan | Instant models | Reasoning models | Input, OpenAI's estimate (Instant) | |---|---|---|---| | Free | 27K tokens | Varies | About 12 pages of text | | Go | 54K tokens | 256K tokens | About 40 pages | | Plus | 54K tokens | 256K tokens | About 40 pages | | Pro | 128K tokens | 400K tokens | About 250 pages | These numbers change with new models, so check OpenAI's [pricing page](https://chatgpt.com/pricing) for your plan. ### Your messages share that budget The window isn't all yours. OpenAI says the space for your input is smaller than the total, because the same window also holds system instructions (including tools and personality), your memories if memory is on, and the model's own work: reviewing information, reasoning and writing the reply. So a conversation that's been going for weeks, with pasted documents, web searches and long answers, fills the window much faster than the message count suggests. ### When the window fills, early details lose out OpenAI doesn't publish exactly how ChatGPT decides what to keep once a conversation outgrows the window. In practice, the symptom is that early details stop being used: the model contradicts a decision from message five, or forgets a rule you set at the start. Even within the window, more isn't always better. Researchers at Stanford and elsewhere found that language models use information at the **beginning and end** of a long input best, and do noticeably worse with information buried in the middle ("Lost in the Middle", Liu et al., 2023). A 200-message chat puts most of what matters in the middle. ### The hard stop is a separate limit The "maximum length" message is a different limit from the context window. OpenAI doesn't publish how long a single conversation can get. When you hit it, the chat stops accepting new messages and you need a new one. The fixes below are about carrying your work over well. ## What to do when a ChatGPT chat gets too long ### 1. Branch before it gets long The best fix is to never let one chat carry everything. When a side question comes up, branch instead of asking it in the main thread. On chatgpt.com, hover over a message, click **More actions** (⋯) and choose **Branch in new chat**. You get a new chat with everything up to that message, and the original stays as it was. The tangent's messages never land in your main thread, so the main thread stays shorter. ### 2. Summarize and restart When a chat is getting heavy (or has already hit the limit), ask for a handover note, then start fresh: 1. In the long chat, ask: *"Write a handover brief for a new chat: the goal, decisions made and why, current state, open questions, constraints and anything I asked you to always do. Be specific, and keep names and numbers exact."* 2. Read the brief and fix anything that's wrong or missing. This step matters, because a summary drops detail. 3. Start a new chat, paste the brief, and continue. If the old chat has already hit the hard limit and won't answer, write the brief yourself from the parts you need, or use fix 5. ### 3. Move the work into a project Projects keep related chats, files and instructions together, and they're available on all free and paid plans. OpenAI says projects have built-in memory: chats can draw on the other chats and files in the same project. 1. Create a project, or open the long chat's menu and choose **Move to project**. 2. Add **project instructions** for the rules that should apply to every chat in it (tone, format, constraints). 3. Upload the reference material as **files** instead of pasting it into messages. The number of files depends on your plan: 5 on Free, 25 on Go and Plus, 40 on Pro, Business, Enterprise and Edu. 4. Start new chats inside the project for each new piece of work. For work you want kept separate from everything else, choose **project-only memory**. Chats in that project can reference each other but nothing outside it. Shared projects always use project-only memory. ### 4. Put lasting facts in memory, not in the thread Some things should apply to every chat: your role, your preferences, a style you always want. Put those in memory so they don't depend on one long thread. ChatGPT's memory has two parts. **Saved memories** are details you've told it to remember ("Remember that I write in British English"). **Reference chat history** lets it draw on past chats. OpenAI is clear that chat history doesn't keep every detail, and recommends saved memories for anything you want ChatGPT to always keep in mind. You'll find both under Memory in Settings. That's also why "continue our last chat" in a new chat often disappoints: it relies on chat-history recall, not on the full old conversation. ### 5. Export the conversation and upload it as reference If you need the old chat's actual content, not a summary: 1. Go to **Settings**, then **Data controls**, then **Export data**. 2. When the export is ready, download the ZIP file, unzip it and find `conversations.json`. Larger exports may be split into numbered conversation files. 3. Start a new chat and upload the file. OpenAI's help center documents this route (for moving conversations to another account): the uploaded file is available as reference in that conversation. The export holds all your chats, so if the file is too big, OpenAI suggests uploading the numbered files separately, splitting the file, or uploading only the conversations you need. Pulling the one conversation you need into its own text file works too. ### 6. Use a model with a bigger window On Go and Plus, the reasoning models list a 256K token window against 54K for Instant. For a long document review or a chat that keeps losing early details, picking a reasoning model in the model picker gives the conversation more room. It doesn't remove the hard length limit. ### 7. Attach long text instead of pasting it OpenAI now turns pastes of more than 10,000 characters into an attachment automatically, to keep big pastes from eating the context window. Do the same on purpose: upload documents as files (in a project, ideally) rather than dropping them into the thread again and again. ## A different shape for long work: a canvas All of the fixes above work around one fact: a chat is a single line, and everything you've ever said in it travels along. Branching helps, but in ChatGPT a branch is just another chat in the list, frozen at the moment you split it. On a canvas, long work becomes several short chats laid out side by side, and **you decide what each chat reads**. That's how [Slashspace](/docs/why-a-canvas) works: - **Each sub-question gets its own chat.** Research, outline, draft and critique live in separate chat nodes, so no single conversation has to carry the whole project. - **Context is wired, not accumulated.** Draw a connection from a source, a note or another chat into a chat, and that chat reads it. [Connected nodes are sent in full](/docs/canvas/connections) with every message. Nothing is summarized or trimmed behind your back, and a chat reads only what's connected to it. - **Branches stay live.** A branch keeps seeing new messages in the chat it came from, and sibling branches don't see each other. You can branch from a single sentence of an answer, not just a whole message. - **You can start clean on purpose.** A chat with nothing connected gets a short summary of the canvas by default (Summarized context), or nothing at all in Isolated context, which is useful for a fresh critique. - **Documents go in once.** Add a long PDF as a [Document node](/docs/nodes/document-node) and connect it only to the chats that need it. They get its full converted text. To be straight about the limits: sending everything in full means the context window still applies. Connect enough big sources and a request can exceed the model's window and fail. The fix is to disconnect what that chat doesn't need or pick a model with a bigger window. A Document node's text over about 500 KB (very roughly 80,000 words) is left out, so split long books into chapters. A canvas doesn't make the window bigger. It makes it obvious what's in it. You can run it on your ChatGPT plan: Slashspace drives the official [Codex command-line tool](/docs/settings/subscription-providers) (and Claude Code, for Claude plans) on your computer, signed in with your own account, and usage counts against your plan. Your own API keys and local models work too. Slashspace is $89/year, starting with a 3-day trial (card required), or $249 one-time lifetime, on Mac, Windows and Linux. [See pricing](/pricing) · [Start a 3-day trial](/checkout/yearly) ## Frequently asked questions ### Why does ChatGPT say my conversation is too long? Each conversation has a maximum length, and long chats with pasted documents, searches and long answers reach it sooner. OpenAI doesn't publish the exact limit. When you see the message, start a new chat and carry your work over with a summary, a project or an exported file. ### How long can a ChatGPT conversation be? OpenAI doesn't publish a maximum conversation length. It does publish context windows, which limit how much the model can read at once: as of October 8, 2026, 27K tokens on Free and 54K on Go and Plus for Instant models, with larger windows for reasoning models and on Pro. ### Does upgrading to Plus fix long conversations? It gives you more room. Plus lists 54K tokens for Instant models against 27K on Free, and 256K for reasoning models. Very long conversations can still outgrow that, so the habits above still help. ### Why does ChatGPT forget things from earlier in the chat? The conversation has likely outgrown what the model can use well. Older details either fall outside the context window or sit in the middle of a very long input, where models are weakest. Restate key decisions, keep them in project instructions or saved memories, or start a new chat with a brief. ### Can memory replace a long conversation? No. Saved memories hold short, lasting facts, and chat-history recall doesn't keep every detail. For the substance of a long project, use a project with files, a handover brief, or an exported conversation. ### Can I continue a ChatGPT conversation in a new chat? Not directly, but you can bring its content along: paste a handover brief, move your work into a project so new chats can draw on it, or upload the exported conversation file as reference. ## Read next - [Understanding tokens and context](/blog/understanding-tokens-and-context) - [Prompt engineering vs context engineering](/blog/prompt-engineering-vs-context-engineerings) - [Linear AI chat apps versus non-linear apps](/blog/linear-ai-chat-apps-versus-non-linear-apps) - [Connections and context (docs)](/docs/canvas/connections) ## How Researchers Use AI to Accelerate Literature Reviews (https://www.slashspace.ai/blog/how-researchers-use-ai-accelerate-literature-reviews) Published: 2026-10-21 AI won't write your literature review, but it can take on the mechanical parts that slow you down: triage, theme finding, contradiction mapping and first-pass synthesis. Here are the workflows that help, and the pitfalls to avoid. Literature reviews are the backbone of serious research. They're also slow, mentally draining and prone to overload. You start with 50 papers, end up with 200, and somewhere in the middle lose track of which findings belong to which themes. AI changes the work, not by replacing your analytical thinking, but by taking on the mechanical parts that bog you down. Used well, it shortens the slowest stages, triage and first-pass synthesis, and makes it easier to see patterns across many papers at once. Here's how researchers use AI to speed up literature reviews in 2026, with workflows you can copy. ## The literature review bottlenecks AI helps with Traditional literature reviews hit predictable walls: **Information overload.** You find 15 relevant papers on Monday. By Friday you've collected 80 more, and you can't hold all the connections in your head. **Context switching.** Reading paper A, then paper B, then trying to remember how both relate to paper C from last week. Every switch costs energy. **Synthesis paralysis.** You understand individual studies but struggle to see patterns across dozens of them. **Citation rabbit holes.** One interesting reference leads to five more papers, which lead to fifteen more. Soon you're three steps away from your original question. **Scattered notes.** Highlights in PDFs, notes in different apps, key quotes spread across documents. When it's time to write, you can't find anything. AI doesn't remove these problems. It makes them manageable by handling the mechanical work while you focus on the thinking. ## Core AI workflows for literature reviews ### 1. Rapid screening and summarization Start by giving AI abstracts and key sections, and ask specific questions: - "What's the main finding of this study?" - "What methodology did they use?" - "How does this relate to [your research question]?" - "Which limitations do the authors mention?" The goal isn't a perfect summary. It's fast triage: deciding which papers deserve a close read and which can be skimmed or set aside. **Practical tip:** write one standard summary template and use it for every paper. Consistent outputs make patterns much easier to spot later. ### 2. Finding themes across sources Once you've processed individual papers, AI is good at spotting patterns across a collection. Give it the summaries of 20 to 30 papers and ask: - "What are the main themes across these studies?" - "Which papers challenge the dominant view?" - "What gaps do you see in this research?" - "How do the methodologies cluster?" This bird's-eye view often reveals connections you missed reading papers one at a time. ### 3. Comparing studies and mapping contradictions AI is good at spotting the methodological differences that can explain conflicting results. Give it papers with opposing findings and ask why they might differ. That saves hours of cross-referencing and often surfaces explanations you'd have missed. ### 4. Working through citation networks Give AI the reference lists of your key papers and ask which works come up most often, which authors keep appearing and where your coverage looks thin. It helps you find foundational papers you missed and understand how your field's ideas developed. Always confirm each reference exists and says what the model claims before you rely on it. ## Advanced techniques for complex reviews ### Compare models to check your findings Different AI models have different strengths and different blind spots. Running the same analysis through two or three models helps you catch claims that only one of them makes. This matters most in systematic reviews, where accuracy outweighs speed. ### Refine in loops Don't expect perfect results on the first pass. Treat AI output as a starting point for sharper questions: 1. Get initial themes from AI 2. Refine them with your domain knowledge 3. Run the analysis again with the updated framing 4. Repeat until the patterns stabilize ### Build bridges to other fields AI can point you to relevant work in neighboring fields that keyword searches miss. Give it papers from your core discipline and ask which related fields might have useful insights. This is especially valuable in interdisciplinary research, where the best ideas often come from unexpected places. ## Putting it into practice ### Organize your files Keep the original PDFs organized by theme or by date. For long papers, decide which sections each question needs (abstract, methods, results, conclusions), so you aren't sending a model far more text than it needs. ### Write research-specific prompts Generic prompts produce generic results. Write prompts for your field: Instead of: "Summarize this paper." Try: "Extract the main hypothesis, sample size, key findings and limitations from this psychology study. Format them as bullet points." ### Manage context in large reviews Literature reviews can involve hundreds of papers, and no model can read them all at once. The answer is to compose context deliberately: give each question only the sources it needs. That's where a tool built for branching, connected conversations helps. Instead of re-pasting papers into a fresh chat every time, you decide which sources and earlier analyses each new question reads. ## Where Slashspace fits in a literature review Ordinary AI chat works fine for questions about one paper. It breaks down when you're juggling many threads and dozens of sources. [Slashspace](/) is an AI canvas that handles this differently. Each paper is a document node. Each research question gets its own chat, and you connect only the papers it should read. A chat reads what's connected to it, and nothing else, so the methods question never leaks into the findings question. Click a chat and every source feeding it lights up, so you can see exactly what an answer drew on. - **Ask about a passage.** Select a sentence in any answer and a branch opens next to it with the passage quoted, so a side question doesn't clutter the main thread. - **Synthesize by theme.** Group papers or finished question chats and connect the group to a new chat. It reads everything inside in full, so you can draft from all of it without copy and paste. - **Check claims across models.** Branch the same question to two or three models and compare the answers side by side on the same papers. Use Slashspace's hosted models, your own API keys, your Claude or ChatGPT subscription through Claude Code or Codex, or local models through Ollama. - **Keep your work.** Every canvas is a folder on your computer. Download it as Markdown when you're ready to write. Every connected paper is sent in full, so a chat wired to many long papers can go past a model's context window. Group papers by theme and synthesize each group first. Slashspace shows which papers each chat drew on, but it isn't a citation manager: ask the model to name the paper behind each point, check it, and cite from your reference manager. See [Slashspace for literature reviews](/for/literature-review). ## Common pitfalls and how to avoid them ### Relying on AI summaries AI summaries are starting points, not endpoints. They miss nuance, misread field-specific terms and sometimes invent details. Always check key claims against the original paper. ### Inconsistent prompts Using different prompts for similar papers makes them hard to compare. Write standard templates and stick to them for the whole review. ### Context pollution Giving AI too much irrelevant material dilutes its answers. Be selective about what each question reads. ### Ignoring methodological differences AI may group studies by similar findings while missing methodological differences that explain why results vary. Always consider methodology when you interpret AI-generated themes. ## The future of AI-assisted research It's still early for AI in academic research. Current tools handle the mechanical parts well but struggle with deep domain knowledge and methodological nuance. The most effective approach combines AI's speed with human expertise. Let AI handle information processing, pattern finding and first-pass synthesis. Keep the critical thinking, the methodological judgment and the final interpretation for yourself. ## Getting started with AI literature reviews Start small. Pick a narrow question with 10 to 15 relevant papers, and develop your prompts and workflow on that set before scaling up. Focus on process before tools. The specific model matters less than a systematic approach to extraction, theme finding and synthesis. Write down what works. Literature reviews come around again, and the workflow you build for one will speed up the next. The goal isn't to automate research thinking. It's to automate research drudgery, so you have more time for the thinking that matters. [See pricing](/pricing) · [Start a 3-day trial](/checkout/yearly) Slashspace is $89/year, starting with a 3-day trial (card required), or $249 one-time lifetime, on Mac, Windows and Linux. ## Read next - [Slashspace for literature reviews](/for/literature-review) - [How to build a research workflow using AI mind maps](/blog/how-to-build-research-workflow-ai-mind-maps) - [Document node (docs)](/docs/nodes/document-node) - [Compare models side by side](/features/compare-models) ## How to Branch a ChatGPT Conversation (Plus Claude and Gemini) (https://www.slashspace.ai/blog/how-to-branch-chatgpt-claude-gemini) Published: 2026-10-08 How to branch a ChatGPT conversation, plus Claude and Gemini, step by step from each app's own instructions, and what to do when branched chats pile up. You're forty messages into a good conversation and a side question pops up. Ask it here and the thread you care about drifts off course. Open a new chat and you lose all the context you built. Branching fixes that. You split the conversation at one point, chase the tangent in the branch, and the original stays as it was. ChatGPT, Claude and Gemini all let you do it now, each in a different way. This guide shows how to branch a ChatGPT conversation, a Claude conversation and a Gemini conversation, using each company's own instructions. Then it covers where built-in branching runs out, and what a canvas does instead. ## What "branching a conversation" means A branch is a copy of a conversation up to a chosen message. Everything after that point is new, and the original thread isn't touched. You'd use it to: - **Chase a tangent** without derailing the main thread - **Try a different prompt** and keep both answers - **Compare approaches** from the same starting context - **Recover** from an answer that went sideways, without starting over ## Quick answer | App | How to branch | Where the branch lives | |---|---|---| | ChatGPT | Under a message: More actions (⋯), then **Branch in new chat** | A new, separate chat | | Claude | **Edit** one of your earlier messages and send | Inside the same chat, with arrows to switch versions | | Gemini | Under a response: **Menu**, then **Branch in a new chat** | A new, separate chat | ## How to branch a conversation in ChatGPT ChatGPT added branching on the web in September 2025. A branch opens as its own chat, so the original stays exactly as it was. **On the web (chatgpt.com):** 1. Sign in and open the conversation you want to branch. 2. Find the message you want to branch from. Everything up to and including it goes into the branch. 3. Hover over the message and click **More actions** (the ⋯ icon). 4. Click **Branch in new chat**. 5. A new chat opens with the conversation up to that point. Ask your new question there. These are the steps in OpenAI's release notes. ChatGPT's interface changes often, so if the menu looks different, look for the same option under the message. **In a project:** branched chats appear in the project next to the original conversation. In a shared project, members can branch a teammate's chat to try a new direction without changing the original. **Tips:** - **Rename the branch** straight away. Branch titles look a lot like the original, and after a few branches the sidebar gets confusing. - **Branch from the last good answer**, not the bad one. The branch includes the message you branch from. - **Your original keeps going.** Anything you add to it afterwards stays there. The branch only knows what came before the split. ## How to branch a conversation in Claude In Claude's chat app, you branch by editing. Anthropic's own guidance: go back to a previous message, click Edit to create a new conversation branch, and you can always return to your original version. **On claude.ai and the desktop app:** 1. Scroll up to **one of your own messages**: the point where you want the conversation to go a different way. 2. Hover over it and click **Edit** (the pencil icon). 3. Change the message, or keep it the same and add your new direction, then send. 4. Claude answers from that point. Everything that followed in the old version is set aside, not deleted. 5. Use the arrows under the edited message to switch between versions. **What to know:** - **You branch from your own messages.** Edit works on what you wrote, not on the middle of Claude's reply. - **Branches stay in one chat.** They don't show up as separate chats in your sidebar. - **Only one version shows at a time.** To compare two versions, flip between them or copy one out. **Using Claude Code?** In the terminal, `/branch` makes a copy of the session so far and switches you into it, leaving the original intact. From the command line, `claude --continue --fork-session` does the same. ## How to branch a conversation in Gemini Google started rolling out branching in Gemini in April 2026, and its help page now covers the web, Android and iPhone. **On the web (gemini.google.com):** 1. Start a chat or open the chat you want to continue. 2. Below the response you want to branch from, click **Menu**. 3. Click **Branch in a new chat**. **On Android or iPhone:** 1. Open the Gemini app. 2. Start a chat or open the chat you want to continue. 3. Below a response, tap **Menu**, then **Branch in a new chat**. **What to know:** - **The two chats are independent.** Deleting the branch doesn't delete the original, and deleting the original doesn't delete the branch. - **You can go back.** Google describes it as branching off from a point and returning to the main conversation to continue it. ## No branch button? The copy-the-context workaround On an app or plan without branching, you can fake it: 1. At the point you want to branch, ask: *"Summarize this conversation so far as a short brief I can paste into a new chat: the goal, key decisions, open questions and any constraints."* 2. Copy the brief. 3. Open a new chat, paste it, and ask your tangent question. It's lossy, because a summary drops detail. But it works anywhere, and the brief makes a decent table of contents for the thread. ## Why "Branch in new chat" isn't enough Built-in branching is a real improvement. But once you branch more than once or twice, the same problems keep coming up. ### 1. There's no map ChatGPT and Gemini branches are separate chats in a list sorted by date. Claude's sit behind arrows inside one chat. Neither shows you **which branch came from where**, so after a week of research you're left with a pile of branched chats and no idea how they relate. ### 2. Each branch is frozen at the split A branch copies the conversation up to one point. If you keep working in the original (you add a source or settle a decision), the branch never finds out. You end up re-pasting context by hand. ### 3. You can't see branches side by side Comparing two branches means switching tabs or clicking arrows and remembering what the other one said. Comparing two *models* on the same context is harder still. ### 4. You can't bring branches back together Branches only split. When two of them reach useful conclusions, nothing combines them except copy and paste. ### 5. You branch whole messages, not ideas Usually the tangent is one sentence in a long answer. Built-in branching copies the whole thread up to that message, and then you have to explain which sentence you meant. ChatGPT and Gemini can also recall your past chats when memory is on. That's useful, but it's recall the model decides on, not a structure you can see or control. ## How branching works on a canvas (Slashspace) This is the problem we built Slashspace for. (You may know it by its old name, RabbitHoles AI.) Instead of a list of chats, you work on an infinite canvas, and **every branch is a node connected to the chat it came from**. ### Branch a whole conversation Type your next message and press **⌘⇧Enter** (Ctrl+Shift+Enter on Windows and Linux), or click the split-arrows button next to Send. A new chat appears to the right and runs your prompt. The original never gets the message. Want an empty branch? Press **⌘B** (Ctrl+B) while typing. ### Branch from one sentence Select any text in an AI answer. An **"Ask about this…"** bubble appears. Type your question and press Enter, and a branch opens with the passage quoted and your question under it. Press Enter without typing to ask the AI to expand on that passage. ### Branches stay live A branch is joined to its parent by a live connection. If you keep talking in the parent, the branch sees those messages the next time you send. Sibling branches can't see each other, and nothing leaks back into the parent. ### Compare models on the same context Press ⌘B once per model, pick a different model in each branch, and send the same prompt. The answers sit side by side, all working from the same context. ### Bring branches back together [Context follows connections](/docs/canvas/connections). Connect two finished branches, plus a PDF or a web page, into a new chat, and it reads all of them in full. That gives you a synthesis without copy and paste. ### Keep the map readable Rename and color branches, group related ones, and press **Tidy Up** to lay everything out left to right, each branch next to its parent. Zoom out and every node becomes a title pill you can click to jump back. ### Use the AI you already have Branches run on whatever you connect: your own API keys, local models through [Ollama](/docs/settings/ollama), or your Claude or ChatGPT subscription. For subscriptions, Slashspace drives the official [Claude Code and Codex command-line tools](/docs/settings/subscription-providers) on your computer, signed in with your own account, and usage counts against your plan. Full details are in the [branching docs](/docs/canvas/branching). ## Branching compared | | ChatGPT | Claude | Gemini | Slashspace | |---|---|---|---|---| | How you branch | Branch in new chat | Edit an earlier message | Branch in a new chat | ⌘⇧Enter, ⌘B, or select text | | Branch from a selected sentence | No | No | No | Yes | | Where branches live | Separate chats | Arrows inside one chat | Separate chats | Connected nodes on a canvas | | See all branches at once | No | No | No | Yes | | Branch sees later parent messages | No | No | No | Yes | | Compare models side by side | No | No | No | Yes | | Combine branches | Copy and paste | Copy and paste | Copy and paste | Connect them into one chat | | Where the official steps apply | Web | claude.ai and desktop app | Web, Android, iPhone | Desktop: Mac, Windows, Linux | To be clear about the trade-off: Slashspace is a desktop app with no web or mobile version yet, so for branching on your phone, the built-in options above are the way to go. ## Who benefits most - **Researchers and students** following several leads from one source, who need to see later which lead came from where - **Writers and marketers** trying three angles on the same brief and picking the best - **Developers** comparing models or approaches on identical context - **Anyone with 50-message threads** who keeps losing the good part If you branch once a week, the built-in buttons are fine. If you branch every session, a map pays off quickly. ## Frequently asked questions ### How do I branch a conversation in ChatGPT? On chatgpt.com, hover over a message, click More actions (⋯), and choose **Branch in new chat**. A new chat opens with everything up to that message, and the original stays unchanged. ### Can you branch a conversation in Claude? Yes, by editing. Edit one of your earlier messages and send it, and Claude creates a new branch from that point. Use the arrows under the message to switch between versions. In Claude Code, use `/branch`. ### Does Gemini have branching? Yes. Below a response, open the Menu and choose **Branch in a new chat**. It works on gemini.google.com and in the Gemini apps for Android and iPhone. ### Does a branch remember the original conversation? It gets everything up to the branch point. In ChatGPT and Gemini, anything you add to the original afterwards doesn't reach the branch. In Slashspace, the branch stays connected to its parent and sees new parent messages. ### Is a branched chat the same as a temporary chat? No. A branch is a normal saved chat that starts with copied context. A temporary chat starts empty and isn't kept in your history. ### Can I merge two branches? Not in ChatGPT, Claude or Gemini, short of copying and pasting. On a Slashspace canvas, connect both branches into a new chat and it reads both. ## The bottom line All three major assistants can branch now. Use **Branch in new chat** in ChatGPT, **Branch in a new chat** in Gemini, and **Edit** in Claude. They're great for the occasional tangent. If you branch a lot, though, you'll soon want to see your branches, compare them side by side and pull them back together. That's what a canvas is for. [See pricing](/pricing) · [Start a 3-day trial](/checkout/yearly) Slashspace is $89/year, starting with a 3-day trial (card required), or $249 one-time lifetime, on Mac, Windows and Linux. ## Read next - [Linear AI chat apps versus non-linear apps](/blog/linear-ai-chat-apps-versus-non-linear-apps) - [Non-linear AI chat: 5x faster](/blog/non-linear-ai-chat-5x-faster) - [Branching in Slashspace (docs)](/docs/canvas/branching) - [Why a canvas?](/docs/why-a-canvas) - [How Slashspace compares](/compare) ## How to Build a Research Workflow Using AI Mind Maps (https://www.slashspace.ai/blog/how-to-build-research-workflow-ai-mind-maps) Published: 2026-10-21 Research used to mean drowning in browser tabs and losing promising threads. Here's a four-stage AI mind map workflow (Scope, Branch, Synthesize, Output) that keeps every thread, source and conclusion connected. Research used to mean drowning in browser tabs, losing promising threads and struggling to connect scattered insights. Linear notes and single chat threads push you down narrow paths, which makes it hard to explore a tangent or see the bigger picture. AI mind mapping changes that. You can branch your research in several directions at once, put different AI models to work on the same question, and organize what you find by meaning rather than by the order you found it. Here's how to build a research workflow that matches how your mind works. ## Why traditional research workflows fall short Most research with AI still follows a linear loop: open a chat, ask questions in sequence, copy the answers somewhere, repeat. That creates four problems. **Context switching kills momentum.** Every jump between your AI chat, your notes and your reference material breaks your concentration. **Linear conversations limit exploration.** Locked into one thread, promising tangents get forgotten or half explored. **Insights end up siloed.** Findings scattered across conversations, documents and apps rarely connect. **Sources get chaotic.** PDFs, web pages and notes live in separate places, so you can't see how they relate. A visual, branching workspace solves these. Every research thread can grow on its own while staying connected to the whole. ## The AI mind map research framework Effective AI research follows four stages: **Scope, Branch, Synthesize, Output.** Each builds on the one before, while leaving room to explore out of order. ### Stage 1: Scope your research territory Start by defining your boundaries and core questions. This isn't about limiting yourself. It's about giving your branches a solid trunk. **Create a central research node.** Begin with a clear statement of what you're investigating, such as "How remote work affects team creativity" or "Blockchain applications in supply chain management". **Name your key dimensions.** What are the three to five main angles you need to explore? These become your first branches. For the remote work example: productivity, collaboration tools, creative processes, team dynamics and differences between industries. **Decide how deep to go.** Do you need a quick brief or a thorough analysis for a major decision? **Write down what you already know.** It prevents repeated work and shows you where the real gaps are. ### Stage 2: Branch into focused conversations This is where AI mind mapping shows its value. Instead of forcing every question through one conversation, give each part of the research its own branch. **Run investigations in parallel.** Create a separate chat for each major dimension, so you can go deep on one area without losing track of the others. **Use the right model for each job.** Different models are good at different things. Use a strong reasoning model for data interpretation, a fast one for quick lookups, and try two models on the questions that matter most. The key is comparing them on the same context rather than starting over in another app. **Follow tangents immediately.** When a conversation reveals an unexpected angle, branch it right away instead of letting it get buried. **Attach sources where they belong.** As you find relevant papers, studies or web pages, connect them to the branches that need them, so sources stay next to the insights they support. ### Stage 3: Synthesize across branches Raw branches are only the beginning. The value comes from connecting insights across areas. **Look for patterns across branches.** Themes that show up in several threads are often your most important findings. **Create synthesis chats.** Start new chats whose only job is to combine insights from different branches. Ask things like "How do the productivity findings relate to the collaboration problems?" or "What changes when we combine the technical limits with the user behavior data?" **Challenge your assumptions.** Use synthesis to test whether your original scope was right. The most valuable insights often come from discovering what you didn't know you didn't know. **Build chains of argument.** Link related insights into a logical sequence. That turns scattered information into understanding. ### Stage 4: Generate focused outputs The last stage turns your research network into deliverables. **Create output branches.** Whether it's a report, a presentation or a recommendation, give each output its own chat. **Draw on your research network.** Connect the specific branches and sources each output relies on, so it's grounded in your whole analysis rather than a few cherry-picked examples. **Iterate with AI.** Use the conversation to tighten arguments, improve clarity and find gaps in your reasoning. **Keep it traceable.** Keep clear links between each output and the research behind it, so you can dig back in when questions come up. ## Putting it into practice ### Start small, then scale Don't overhaul your whole research process at once. Start with one project and get comfortable with branching. **Week 1:** practice creating three to five branches for a single topic, keeping clear links between them. **Week 2:** compare models. Put the same question to two or three models and notice how their answers differ. **Week 3:** add documents and web pages to your branches, and practice deciding which chats should read which sources. **Week 4:** focus on synthesis. Create chats whose only job is to connect insights across your network. ### Organize for long-term value Research compounds when it's well organized. **Use consistent names.** Label chats so they still make sense weeks later, with the project and the focus in the title. **Reuse what works.** When a canvas layout works well, duplicate it as the starting point for a similar project. **Keep a library.** Keep useful research canvases where you can find them. Today's background research is often tomorrow's starting point. **Note what works.** Workflows improve through reflection. Keep a short note of what helped and what didn't. ### Share your reasoning, not just your conclusions Research is more useful when others can see how you got there. **Share the relevant part.** Instead of sending everything, share the threads that matter for a specific decision. **Build on each other's work.** When a colleague shares research, extend it with your own branches instead of starting from scratch. **Plan handoffs.** Agree how research moves between people or project phases, so context isn't lost. ## Advanced techniques ### Compare models on purpose Different models have different strengths, and their disagreements are informative. **Use reasoning models** for interpretation, analysis and logic. **Use other models** for brainstorming, alternative perspectives and unexpected connections. **Use specialized or local models** for technical domains or sensitive material. **Combine their answers** in a synthesis branch to get a fuller picture than any single model gives. ### Keep sources current Research goes stale. Build in ways to bring in new information. **Keep monitoring branches** for topics where new information keeps arriving. **Refresh key canvases** with new data or perspectives on a schedule. **Evaluate new sources quickly** with a standard set of credibility and relevance questions. ### Analyze your research network Big projects produce complex networks of insight. **Map dependencies.** Note which findings depend on others and which stand alone. **Find gaps.** Look for thin branches and weak connections. **Spot high-impact nodes.** Some insights affect many areas. Prioritize those. ## What to look for in an AI mind mapping tool for research The principles work anywhere, but the tool shapes how easy they are. **Branching.** You should be able to fork a conversation and keep clear links between the branches. **Model choice.** Comparing models on the same context should be easy. **Sources as context.** You should be able to attach documents, web pages and other material to a conversation and control which conversations read them. **Spatial organization.** Seeing the whole network helps you navigate and spot relationships. [Slashspace](/) is built around exactly these. Every chat is a node on an infinite canvas, and you can branch any chat, or a single sentence of an answer, into a new one. Each chat has its own model: Slashspace's hosted models, your own API keys, your Claude or ChatGPT subscription through Claude Code or Codex, Cursor, or local models through Ollama. PDFs, Office files, web pages, YouTube and social posts become nodes you connect to the chats that need them. Orchestrator mode can even lay out a starting canvas from a research goal. As a canvas grows, groups, Stacks and Tidy Up keep it readable, and `⌘K` search finds any node. Canvases are plain folders on your computer, and you can download one as Markdown or JSON to take your research elsewhere. To show someone how your research came together, publish the canvas as a Replay (beta) and choose exactly which parts they see. ## Common pitfalls and how to avoid them ### Branching without purpose A branch for every small tangent creates chaos, not clarity. **Fix:** set a simple rule for when a question deserves its own branch. Save branches for genuinely new angles. ### Skipping synthesis Collecting insights without connecting them produces impressive-looking research with no conclusions. **Fix:** schedule synthesis sessions and treat connecting insights as seriously as gathering them. ### Losing track of sources As branches multiply, it's easy to forget where an insight came from. **Fix:** keep each source connected to the branches that used it, and note in your synthesis where each claim came from. ### Analysis paralysis Infinite branches can mean endless research and no decision. **Fix:** set deadlines and decide on the output up front. Use time limits to force synthesis. ## How to tell if your workflow is working Ask yourself these questions every few projects: **Time to insight:** how quickly do you get from a question to an understanding you can act on? **Source variety:** are you drawing on several kinds of sources and perspectives? **Unexpected connections:** how often do you find relationships between different parts of your research? **Output quality:** are your research-based decisions and recommendations getting better? **Reuse:** how often do you build on earlier research instead of starting over? ## Building your research advantage Good research has always been about asking better questions, not just finding more answers. AI mind mapping amplifies that by letting you follow several lines of questioning at once while keeping the context each one needs. The framework here (Scope, Branch, Synthesize, Output) gives you structure without taking away the freedom that makes research creative. Try it on your next project. Write your central question, branch into the key dimensions, and see how a canvas turns scattered information into connected insight. [See pricing](/pricing) · [Start a 3-day trial](/checkout/yearly) Slashspace is $89/year, starting with a 3-day trial (card required), or $249 one-time lifetime, on Mac, Windows and Linux. ## Read next - [How researchers use AI to accelerate literature reviews](/blog/how-researchers-use-ai-accelerate-literature-reviews) - [How to use AI to explore any topic deeply](/blog/how-to-use-ai-explore-any-topic-deeply) - [Slashspace for research](/for/research) - [Branching in Slashspace (docs)](/docs/canvas/branching) ## How to Use AI to Connect Ideas Across Different Topics (https://www.slashspace.ai/blog/how-to-use-ai-connect-ideas-across-topics) Published: 2026-10-21 Your working memory can only hold a few ideas at once, which makes connections between fields hard to spot. AI can draw on many domains at once. Here's how to use it to find cross-domain connections, and how to check them. Your mind is great at recognizing patterns, but it hits a wall when it has to juggle many ideas at once. Working memory holds only a few things at a time, which makes it surprisingly hard to spot connections between different fields, even when those connections could unlock a new idea. AI doesn't have that limit in the same way. It can draw on material from many domains in one answer and suggest relationships you might not notice for years. The real question isn't whether AI can connect ideas. It's how to use that ability deliberately, and how to tell a useful connection from a merely clever one. ## Why cross-domain connections matter New ideas often come from intersections. George de Mestral invented Velcro after studying how burrs clung to his dog's fur. August Kekulé said the ring structure of benzene came to him in a daydream of a snake seizing its own tail. Biomimicry, as a whole field, borrows designs from nature to solve engineering problems. These moments happen when someone spots a pattern from one field that solves a problem in another. Making those connections takes broad exposure to different ideas and the room to hold them side by side, which is where AI can help. ## Core techniques for connecting ideas with AI ### Start with analogies Instead of asking "How do I solve this marketing problem?", ask "Which principles from biology, physics or game design might apply here?" The shift invites the model to draw on other fields instead of staying inside familiar ones. You might find that customer retention looks like ecosystem stability, or that a pricing problem resembles a classic auction design. The key is reframing the challenge so it invites cross-domain thinking. ### Bridge a concept Take a core concept from your area and explore where it shows up elsewhere. Working on team dynamics? Ask how feedback loops, emergence or network effects appear in ecology, city planning or computer science. Be specific about the concept and open about where it leads. This reveals unexpected applications and fresh angles on familiar problems. ### Transfer constraints Every field works within limits. Ask AI to describe a constraint in your domain, then explore how other fields handle similar limits. How do birds deal with weight limits? How do software systems manage scarce resources? How do jazz musicians work inside a fixed structure? These constraint and solution pairs often transfer well, because they come with frameworks that have already been tested. ## Frameworks for making connections ### The three-domain method Pick your target area, then two unrelated fields. Ask AI which principles from those fields might apply to your challenge. The tension between perspectives often sparks something new. Designing a learning platform? Explore restaurant operations (service flow, guest experience) and forest ecosystems (growth patterns, how resources spread). The combination can reveal ideas neither field offers alone. ### Climb the ladder of abstraction Start with a specific problem, then ask AI to describe it in more and more general terms. At a high enough level, look for that general pattern in completely different fields. A customer support workflow becomes "a system for receiving requests and routing them efficiently". At that level it connects to air traffic control, immune responses and content moderation. ### Mine history Ask AI how similar challenges were handled in other cultures and eras. Old solutions often contain principles that still apply. How did ancient trade networks handle trust? How did medieval guilds pass on knowledge? How have communities managed shared resources? These patterns often outlive their original setting. ## Advanced strategies ### Compare models Different AI models make different kinds of connections. Put the same prompt to several models and you'll often get different relationships: one leans on scientific analogies, another on history or art. Compare their answers and look for patterns that show up across several of them. Ideas that several models reach independently are worth a closer look; ideas only one reaches are worth checking twice. ### Layer context Build connections step by step by adding material from different fields one layer at a time. Start with your core concept, then bring in a new context and ask what connections appear with each addition. This mirrors how insights usually emerge: not from a single flash, but from perspectives piling up until a pattern becomes clear. ### Study successful crossovers Look at innovations that already bridged fields, and ask AI to analyze how the connection was made. Understanding how others made the leap gives you templates for your own. Look at how biomimicry works, how design thinking moved across industries, or how mathematical ideas found their way into art. Those patterns become reusable frameworks. ## Getting past common barriers ### Breaking out of your field We tend to think inside professional boundaries, as marketers, engineers or artists. AI isn't tied to one identity, which makes it useful for stepping outside yours. Ask AI explicitly to ignore field boundaries. Describe your problem in terms of how it works rather than in your industry's vocabulary. That removes barriers that hide useful connections. ### Managing overload AI can produce far more possible connections than you can use. You need filters. Not every connection is useful, and chasing too many directions scatters your thinking. Favor connections that lead to action, challenge an assumption or offer a clear advantage. Quality beats quantity. ### Checking whether an idea transfers A principle that works in one field won't automatically work in another. Use AI to examine the conditions that make the principle work where it came from. Then check whether those conditions exist in your field, what you'd need to adapt, and what could go wrong. This step separates a useful insight from an expensive mistake. ## Tools and habits for systematic connection-making ### Keep conversations going, and branch them Good cross-domain thinking happens over extended conversations, where each insight builds on the last. A single prompt rarely gets you there. Use a tool that lets you branch a conversation in different directions while keeping each branch linked to the original question. That matches how creative thinking really works: sustained exploration, not isolated queries. ### Bring in sources Give AI documents, articles or data from the fields you're connecting, instead of relying only on what it learned in training. This matters most for niche fields and recent research the model may know little about. ### Map it visually AI is good at proposing connections. People are good at spatial reasoning. Seeing the connections laid out helps you spot higher-level patterns and clusters that a text conversation hides. ## How to judge a connection Use a few criteria to evaluate what AI suggests: **Novelty:** does it reveal something new, or restate what you already knew? **Actionability:** can you do something with it, or is it only interesting? **Transferability:** do the underlying principles really carry over between fields? **Impact:** if it holds, how much could it change your approach or results? These help you focus on promising connections and avoid unproductive rabbit holes. ## Making connections stick Finding a connection is only the first step. The valuable ones become part of how you think and work. Write down promising connections, test them with small experiments, and keep them somewhere you'll see them again. Over time you'll build a personal collection of patterns and frameworks that make the next insight come faster. ## Where Slashspace fits [Slashspace](/) is an AI canvas made for this kind of exploration. Every chat is a node you can branch, so each field you're drawing from can get its own conversation while staying linked to your core question. Select one sentence of an answer to ask about just that idea. Branch the same prompt once per model to compare how Claude, GPT, Gemini or a local model connects the ideas, side by side. Sources from each field (PDFs, web pages, YouTube and social posts, notes) become nodes you connect to the chats that should read them, so each line of thinking reads only what it needs. When you're ready, connect the most promising branches to one synthesis chat and ask what they have in common. With Quick Window in Slashspace 4.0, you can even send a stray idea from any app on your Mac straight onto the canvas. The future belongs to people who think across fields, and AI gives that thinking a scaffold. [See pricing](/pricing) · [Start a 3-day trial](/checkout/yearly) Slashspace is $89/year, starting with a 3-day trial (card required), or $249 one-time lifetime, on Mac, Windows and Linux. ## Read next - [How to use AI to explore any topic deeply](/blog/how-to-use-ai-explore-any-topic-deeply) - [What is a knowledge graph and how can AI help you build one?](/blog/what-is-knowledge-graph-ai) - [Compare models side by side](/features/compare-models) - [Branching in Slashspace (docs)](/docs/canvas/branching) ## How to Use AI to Explore Any Topic Deeply (A Step-by-Step Guide) (https://www.slashspace.ai/blog/how-to-use-ai-explore-any-topic-deeply) Published: 2026-10-21 Most people use AI like a search box: ask, skim, move on. Real research is a conversation with a topic. Here's a step-by-step workflow for using AI for depth, not just speed. ## The problem with shallow AI research Most people use AI like a search box: ask a question, skim the answer, move on. That works for quick facts. It fails when you need to understand something complex. Real research isn't one question. It's a conversation with a topic. You start somewhere, find gaps in what you know, chase connections, hit contradictions, step back for perspective, then go deeper. That kind of thinking doesn't fit in a single chat exchange. AI can be a powerful research partner when you use it well. Most people never get past surface-level answers because they don't have a workflow that supports deeper exploration. This guide gives you one. Whether you're investigating a business opportunity, learning a hard subject, preparing for a big decision or following plain curiosity, here's how to use AI for depth, not just speed. ## Step 1: Start with a map, not a question Your instinct is to open a chat and fire off your main question. Don't, at least not yet. Spend five minutes working out what you're really after. What do you already know? What exactly are you trying to understand? What would a useful answer look like? Vague curiosity gets vague results. "Tell me about climate change" returns an encyclopedia summary. "Help me understand why carbon capture projects so often struggle to scale despite heavy investment" gives you something to work with. **Before your first prompt, try this:** 1. Write your topic in one sentence 2. Note what you already believe about it 3. Name what specifically confuses or intrigues you 4. Ask what you'd do differently if you understood it better That last question matters most. It ties your research to something real: a decision, a project or a belief worth testing. Research without stakes rarely goes deep. ## Step 2: Use AI to build your first mental model Your opening conversation should be about structure before content. Ask AI how experts think about the topic before asking what they think. **Instead of:** "What is quantum computing?" **Try:** "Before we explore quantum computing, help me understand how experts organize this field. What are the main areas? Which foundational concepts do I need first? What misconceptions do newcomers usually bring?" This shows you the shape of the field, shows you what you don't know yet, and gives you a frame for everything that follows. Think of it as sketching a map. You're not visiting every place yet. You're finding out which places exist. ## Step 3: Branch into subtopics on purpose This is where most AI research falls apart. You have a good first conversation, then keep piling everything into the same thread. The conversation gets unwieldy, the model is reading a mix of unrelated tangents, and you lose track of which angles you've covered. A better approach: branch deliberately. When a subtopic deserves its own exploration, give it its own conversation instead of layering it into the original thread. Say you're researching the future of remote work and your first chat surfaces three threads: management challenges, real estate effects and productivity research. Each deserves focused attention. They'll go deeper and stay cleaner as separate conversations. This is what [Slashspace](/) is built for. Instead of one endless chat, you work on an infinite canvas where every chat is a node. Send a follow-up into a new branch, or select a single sentence in an answer and ask about it, and the original thread stays exactly as it was. Each branch starts with everything its parent knew, and sibling branches never see each other. See [branching](/features/branching). **A practical branching structure:** - **Main node:** your central question and first framework - **First-level branches:** major subtopics worth exploring on their own - **Second-level branches:** specific claims, contradictions or rabbit holes from those subtopics - **Synthesis node:** a separate chat where you bring threads together and test your conclusions ## Step 4: Prompt for depth Research quality depends heavily on your prompts. These strategies reliably produce deeper, more useful answers. ### Ask for the strongest counterargument Whatever position you're forming, ask AI for the strongest case against it. Not weak objections: the best version of the opposing view. *"I'm leaning toward believing open-source AI models will catch up with proprietary ones within five years. What's the strongest argument against this? Give me the version that would make a thoughtful expert pause."* This keeps you honest and often surfaces the most important nuances. ### Ask what experts disagree about Consensus is where topics settle. The interesting part lives in the disagreements. *"What do serious researchers in this field disagree about? Not fringe views. What are the open questions where informed people reach different conclusions?"* This reliably surfaces the live edges of a topic. ### Ask for concrete examples and failures Abstract explanations are easy to generate and hard to remember. Examples stick. *"Give me three specific examples of this working well, and one or two cases where it failed or produced unexpected results."* Failure cases are especially valuable. They reveal hidden assumptions and edge conditions that success stories skip. ### Ask AI where it might be wrong AI models can be overconfident. Asking them to flag uncertainty helps you calibrate. *"Where is your knowledge of this topic likely to be incomplete, outdated or unreliable? What should I verify independently?"* You won't always get a perfect answer, but you'll often get a useful one, and it builds the habit of treating AI output as a starting point, not the final word. ### Use the "explain to an expert" prompt Most AI explanations default to a general audience. If you want depth, say you can handle it. *"Explain this assuming I know the basics and have followed this space for years. Skip the intro and go straight to the nuances."* ## Step 5: Bring real sources into the conversation A model's knowledge has a cutoff date and doesn't include niche or private material. For serious research, bring outside sources into the conversation. The workflow is simple: 1. Ask AI which sources would be most valuable (specific journals, researchers, organizations or report types) 2. Find those sources yourself 3. Bring them into your AI conversation as context When you add a research paper, an earnings report, a policy document or a news article, the conversation becomes grounded and specific. You're working from actual material instead of the model's general training. In Slashspace, sources are nodes on the canvas: PDFs and Office files, web pages, YouTube videos, TikTok, Instagram, X and LinkedIn posts, images and notes. Connect a source to a chat and that chat reads it in full. Chats you don't connect it to never see it, so each line of research reads only what it should. See [connections and context](/docs/canvas/connections). A useful pattern: give each major source its own review chat. Ask AI to summarize the key claims, flag methodological concerns and list the questions it raises. Then branch from that chat into whatever threads the source opens up. ## Step 6: Use different models for different jobs Models have different strengths. Using one model for your whole research process is like using one tool for every job in a workshop. Some models are better at structured analysis, others at creative synthesis or technical depth. Over time you'll learn which you trust for what. Put the same question to several models and look at where they disagree. Disagreement often marks genuinely uncertain or contested ground. Slashspace makes this practical. Every chat node has its own model, so you can branch a question once per model and read the answers side by side, all from the same context. You can use Slashspace's hosted models, your own API keys, your Claude or ChatGPT subscription through Claude Code or Codex, Cursor, or local models through Ollama. See [compare models](/features/compare-models). ## Step 7: Synthesize deliberately Research without synthesis is just collecting. The goal is to understand, not to accumulate. Build synthesis into the process on purpose. **A synthesis prompt that works:** *"I've been researching [topic] from several angles. Here's what I currently believe: [your current model]. Here's what I'm still unsure about: [open questions]. Here's where I've hit contradictions: [tensions]. Help me stress-test this. What am I probably getting wrong? What am I missing? What would change my view?"* This treats AI as a thinking partner rather than an answer machine. It takes longer to write and produces much better output. On a canvas, a synthesis chat is just a chat with your key branches connected to it, so it reads exactly the conversations you choose and nothing else. **Questions worth answering after any deep research session:** - What's the most important thing I learned that I didn't know before? - Which assumption did I bring in that now looks wrong or incomplete? - Which question am I still missing a good answer to? - What would have to be true for my current conclusion to be wrong? - What's the most important next step, in research or in action? Writing short answers to these consolidates what you've learned and makes it stick. ## Step 8: Organize your research so you can use it Deep research only pays off if you can find and apply it later. A great insight buried in conversation number 47 is effectively lost. Build in light organization as you go: - **Name chats clearly.** Not "Chat 1" but "Remote work: productivity research" or "Carbon capture: why projects stall". - **Keep a running synthesis note** with key insights, open questions and changed beliefs. - **Mark the threads worth returning to.** Not everything deserves equal attention. Spatial organization helps. When your research lives on a canvas instead of in a list of chat histories, you can see how threads relate, spot gaps and jump back to a conversation without scrolling through old messages. In Slashspace you can group related nodes, color-code a line of thinking, fold finished work into Stacks, search the canvas with `⌘K`, and download the whole canvas as Markdown or JSON. ## A sample research workflow in practice Here's how the framework plays out on a real kind of question. **Scenario:** you're deciding whether to launch a product in a new market: B2B software for construction companies. **Node 1, mental model:** ask AI to map the construction software landscape. What are the major categories? Who are the incumbents? Which problems stay unsolved? **Node 2, customer branch:** go deep on the buyers. Who makes software decisions? What does the buying process look like? Why might they distrust software vendors? **Node 3, competitor branch:** analyze specific competitors. What do they do well? What do customers complain about? Where are the positioning gaps? **Node 4, source review:** add industry reports, public user reviews and case studies as sources, and ask AI to pull out the key signals. **Node 5, counterargument branch:** ask AI for the strongest case against entering this market. What would a skeptical investor say? **Node 6, synthesis:** connect the branches that matter and bring everything together. What's your current hypothesis? What are you least sure of? What would change your mind? Six focused conversations, clearly organized, each going deep on its own angle. That's a research process, not just a chat session. ## Go deeper, not just faster AI makes research faster. The bigger opportunity is going deeper than you could before: following more threads, testing more assumptions, combining more sources, and doing it without losing track of where you've been. That takes a workflow. It means treating AI as a thinking partner, not a search engine. It means branching, not just scrolling. And it means organizing your research on a canvas, so you can see what you're building. If you're ready to do serious research with AI, Slashspace is built for exactly this. The infinite canvas, branching chats, any model you like and sources wired into each chat make the workflow in this guide feel natural. [See pricing](/pricing) · [Start a 3-day trial](/checkout/yearly) Slashspace is $89/year, starting with a 3-day trial (card required), or $249 one-time lifetime, on Mac, Windows and Linux. ## Read next - [How to build a research workflow using AI mind maps](/blog/how-to-build-research-workflow-ai-mind-maps) - [How to use AI to connect ideas across different topics](/blog/how-to-use-ai-connect-ideas-across-topics) - [Slashspace for research](/for/research) - [Why a canvas?](/docs/why-a-canvas) ## Linear AI Chat Apps Versus Non-Linear Apps (https://www.slashspace.ai/blog/linear-ai-chat-apps-versus-non-linear-apps) Published: 2025-12-12 A practical guide to understanding the differences between linear and non-linear AI chat apps ## Linear Chat Apps These are simple chat interfaces. One window, one isolated piece of context. In linear chat, your “context engineering” is mostly accidental. The interface decides what the model sees: typically the running conversation plus some hidden system instructions plus whatever retrieval the product adds. So when a user says: “It’s giving weird answers,” a lot of times the real issue isn’t the prompt, it’s the context. Some earlier line, assumption, or tangent is still in the thread and it’s quietly shaping everything that comes after. This is why in linear chat apps you see behaviors like: * The model anchoring on an old definition you no longer mean * The model referencing something from 30 messages ago that is no longer relevant * The user having to say “ignore everything above” (which sometimes works, sometimes doesn’t) * The best workaround being: “start a new chat” | Advantages | Disadvantages | |---|---| | Simple | Lack of control | | Low cognitive overhead (just “talk to it”) | Context is implicit and always-on (hard to “turn off” parts of the conversation) | | Great for quick questions and straightforward tasks | Long threads get noisy: irrelevant history gets dragged forward | | Easy to ship and easy to understand as a product | Hard to reproduce outcomes because context drift accumulates over time | | Natural conversational feel | Hard to compare alternatives (you end up opening multiple tabs/threads) | | Works well when the user’s intent is stable | Hard to run parallel lines of thought without mixing them | That tradeoff with Linear chat apps is fine when we need quick answers, but it gets painful when we’re trying to work along with AI to arrive something we need. ## Non Linear Chat Apps Non-linear chat interface like RabbitHoles AI allow you to plug and unplug context. These interfaces offer users superior control to engineer context than linear chat apps. Instead of the single window always dragging the entire thread forward, you can treat context more like a set of building blocks. This makes context engineering visible. It becomes a deliberate act instead of something that happens passively as the chat scrolls. You can think of it like this: * Linear chat = one timeline, one “bag” of context that keeps growing * Non-linear chat = multiple branches + explicit context modules that you can attach/detach ### Why non-linear matters (practically) Most real work is not one clean conversation. It’s more like: * Explore → refine → compare → decide * Draft → critique → revise * Research → synthesize → write * Brainstorm → shortlist → expand | Advantages | Disadvantages / tradeoffs | |---|---| | **Context control:** You decide what is in scope for a response | **More complexity:** More controls means more decisions for the user | | **Less drift:** You can isolate experimental branches from the “main” line | **Higher cognitive load:** You’re not just chatting, you’re managing context | | **Parallel thinking:** You can explore multiple approaches side-by-side | **Onboarding required:** It’s not immediately obvious why branching and context modules matter | | **Better iteration:** You can keep source material stable while swapping prompts, or vice versa | **Potential over-engineering:** For simple tasks, it can feel like too much machinery | | **More reproducible outputs:** Since the context set is explicit, results are easier to recreate | | | **Better collaboration (often):** context modules can be shared, reused, and standardized | | ## TL;DR * Linear chat apps = one continuous thread where context accumulates automatically. * Pros: simple, low effort. * Cons: low control, context drift, hard to run parallel ideas, often you need “start a new chat.” * Non-linear chat apps = branching + modular context you can attach/detach (“plug/unplug”). * Pros: better context control, less drift, easier iteration and comparison, more reproducible. * Cons: more complex, higher cognitive load, can be overkill for simple tasks. ## Why Non-Linear AI Chat Is 5x Faster Than Linear Chat (Backed by Research) (https://www.slashspace.ai/blog/non-linear-ai-chat-5x-faster) Published: 2026-05-09 Linear AI chat forces you to re-explain yourself every session. Non-linear AI chat on a canvas keeps your context alive and connected. Here's the peer-reviewed research showing why that makes you up to 5x faster. You've been there. You open ChatGPT, start a fresh chat, and spend the first five minutes re-explaining your project, your constraints, and what you already tried. Then you do it again tomorrow. And the day after. That re-entry tax adds up fast. And it's just one of several places where linear AI chat quietly burns your time. We built [Rabbitholes](/) as a non-linear AI chat canvas because we believed spatial thinking with AI should be faster. But we didn't want to just *believe* it. We wanted to measure it. After digging through four separate bodies of peer-reviewed research, we landed on a compounded estimate: working with AI on a canvas is roughly **5x faster** than a traditional linear chat window for complex knowledge work. Here's how we got there. ## You waste 30-40% of your time re-explaining context Every time you start a new chat in a [linear AI app](/blog/linear-ai-chat-apps-versus-non-linear-apps), you're starting from scratch. Who you are, what you're building, what you've already explored. All of that gets typed out again. In a non-linear AI canvas like Rabbitholes, context flows between connected nodes. Your research from yesterday is still there, linked to the conversation you're having today. You pick up where you left off instead of rebuilding from zero. **What the research says:** Gloria Mark, Victor Gonzalez, and Justin Harris at UC Irvine studied how knowledge workers handle interruptions. They found it takes an average of **23 minutes and 15 seconds** to fully return to a task after a disruption. Their 2005 study remains one of the most cited papers on the cost of context loss in knowledge work. For people using linear chat tools, that disruption happens every single session. You close the tab, you lose the thread. A spatial canvas that preserves and connects your context cuts this overhead to near zero. **Estimated speed gain: 1.5 to 2x** > Mark, G., Gonzalez, V. M., & Harris, J. (2005). *No task left behind? Examining the nature of fragmented work.* Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI '05), 321-330. > [https://doi.org/10.1145/1054972.1055017](https://doi.org/10.1145/1054972.1055017) ## Context switching between AI threads kills your focus This one goes beyond just re-entering context. Jumping between different chat windows, tabs, and tools has a real cognitive cost, even when the switch only takes a few seconds. Think about a typical research session. You've got one chat for brainstorming, another for fact-checking, maybe a third where you're drafting. Every jump between them fragments your attention. You lose your train of thought. You forget what you were about to ask. On a canvas, all of those threads sit side by side. You glance between them instead of navigating. The spatial layout keeps everything visible and accessible without forcing a full context switch. **What the research says:** Monsell (2003) reviewed decades of task-switching experiments and found that switch costs persist even when people have plenty of time to prepare. The "residue" from your previous task lingers and reduces your performance on the next one. Rubinstein, Meyer, and Evans (2001) showed similar results: switching between tasks increases both time and error rates, even for simple activities. When your AI conversations live on a canvas, you reduce the number of hard context switches. You scan instead of switch, and that makes a real difference over a multi-hour session. **Estimated speed gain: 1.5 to 2x** > Monsell, S. (2003). *Task switching.* Trends in Cognitive Sciences, 7(3), 134-140. > [https://doi.org/10.1016/S1364-6613(03)00028-7](https://doi.org/10.1016/S1364-6613(03)00028-7) > Rubinstein, J. S., Meyer, D. E., & Evans, J. E. (2001). *Executive control of cognitive processes in task switching.* Journal of Experimental Psychology: Human Perception and Performance, 27(4), 763-797. > [https://doi.org/10.1037/0096-1523.27.4.763](https://doi.org/10.1037/0096-1523.27.4.763) ## Spatial layouts free up your working memory When information is laid out in space rather than buried in a scroll, your brain offloads the organizational work to your eyes. You see relationships between ideas instead of trying to hold them all in your head. This is the same reason sticky notes on a wall work better than a bullet list for brainstorming. Physical layout carries meaning. Position, distance, and grouping all communicate structure without requiring you to remember it. **What the research says:** Nesbit and Adesope (2006) ran a meta-analysis across 55 studies comparing concept maps and knowledge maps to text-based formats. Spatial knowledge representations led to significantly better retention and transfer, with effect sizes between 0.4 and 0.8 standard deviations depending on the task. Sweller's cognitive load theory (1988, 2011) explains the mechanism. Spatial layouts reduce what he calls "extraneous" cognitive load: the mental effort you spend organizing and navigating information instead of actually thinking about it. An [AI canvas](/blog/what-is-rabbitholes-ai) turns that overhead into structure you can see. **Estimated speed gain: 1.3 to 1.5x** > Nesbit, J. C., & Adesope, O. O. (2006). *Learning with concept and knowledge maps: A meta-analysis.* Review of Educational Research, 76(3), 413-448. > [https://doi.org/10.3102/00346543076003413](https://doi.org/10.3102/00346543076003413) > Sweller, J. (2011). *Cognitive load theory.* In J. P. Mestre & B. H. Ross (Eds.), Psychology of Learning and Motivation (Vol. 55, pp. 37-76). Academic Press. > [https://doi.org/10.1016/B978-0-12-387691-1.00002-8](https://doi.org/10.1016/B978-0-12-387691-1.00002-8) ## You remember more when you think spatially (dual coding) A canvas doesn't just display text. It encodes your information in two channels at once: the words in each conversation, and the spatial position, color, and connections between nodes. Your brain processes these through separate systems, and that double encoding makes recall significantly faster. You probably already do this intuitively. When someone asks where you read something, you often remember *where on the page* it was before you remember the exact words. That's spatial memory at work. A canvas makes this happen naturally across all your AI conversations. **What the research says:** Allan Paivio's dual coding theory (1986) showed that information encoded in both verbal and spatial formats is recalled 2 to 3x better than information encoded in just one format. The effect holds across free recall, recognition, paired-associate learning, and problem-solving tasks. More recent neuroimaging studies confirm the underlying mechanism: verbal and spatial processing light up distinct cortical networks. Engaging both creates richer memory traces with more paths back to the information when you need it. **Estimated speed gain: 2 to 3x recall improvement** > Paivio, A. (1986). *Mental representations: A dual coding approach.* Oxford University Press. > [https://doi.org/10.1093/acprof:oso/9780195066661.001.0001](https://doi.org/10.1093/acprof:oso/9780195066661.001.0001) > Mayer, R. E. (2009). *Multimedia learning* (2nd ed.). Cambridge University Press. > [https://doi.org/10.1017/CBO9780511811678](https://doi.org/10.1017/CBO9780511811678) ## How these four factors compound These gains don't exist in isolation. They stack on top of each other during a real work session. | Factor | Speed gain | What changes | |---|---|---| | No more re-explaining context | 1.5 to 2x | You pick up where you left off | | Less context switching | 1.5 to 2x | You scan instead of navigate | | Spatial cognitive offloading | 1.3 to 1.5x | The layout carries the structure | | Dual coding recall boost | 2 to 3x | You find things faster | **Conservative estimate (lower bounds):** 1.5 x 1.5 x 1.3 x 2.0 = **5.85x** **Aggressive estimate (upper bounds):** 2.0 x 2.0 x 1.5 x 3.0 = **18x** We call it **5x** because it sits at the conservative end of the compounded range. For complex, multi-threaded knowledge work like research, planning, and iterative problem-solving, we're confident that number holds up. For quick single-question lookups, the difference is smaller. But for deep, multi-hour sessions where you're working across dozens of connected threads? The real advantage likely exceeds 5x. ## Who benefits most from non-linear AI chat? Not everyone needs a canvas. If you're asking ChatGPT to write a quick email, linear chat is fine. But if your work looks like any of these, non-linear AI chat pays off fast: - **Researchers** building literature reviews across dozens of papers - **Product managers** exploring competing approaches to a feature - **Students** studying complex topics with lots of interconnected ideas - **Writers** doing deep research before drafting long-form content - **Consultants** juggling multiple workstreams with different contexts - **Engineers** debugging across multiple files, logs, and documentation The common thread is complexity. Whenever your thinking branches, loops back, or runs in parallel, a [spatial AI workspace](/docs/why-a-canvas) outperforms a single chat window. ## Frequently asked questions ### What is non-linear AI chat? Non-linear AI chat lets you work with AI across multiple connected conversations on a spatial canvas, instead of a single scrolling thread. You can branch, connect, and arrange your AI conversations visually, which gives you control over what context the AI sees and how different ideas relate to each other. ### How is an AI canvas different from regular ChatGPT? ChatGPT and similar tools use a linear chat format: one message after another in a single thread. An AI canvas like Rabbitholes places each conversation as a node on a 2D workspace. You can connect nodes to share context between them, run parallel lines of thinking, and see your entire research landscape at a glance. ### Is non-linear AI chat actually faster? Based on four bodies of peer-reviewed research covering context switching, cognitive load, and dual coding theory, the compounded speed gain for complex knowledge work sits around 5x at the conservative end. The gains are smaller for simple tasks and larger for deep, multi-session research. ### Do I need to be technical to use a non-linear AI tool? No. If you've used sticky notes on a wall or a whiteboard for brainstorming, you already understand the concept. Tools like [Rabbitholes](/) are designed to feel intuitive: you create nodes, connect them, and chat with AI in each one. ## The bottom line Linear chat was built for conversations. But your thinking isn't a conversation. It's a web of connected ideas, branching questions, and parallel threads. When your tools match the shape of your thinking, you stop fighting the interface and start moving at the speed of your ideas. [See pricing](/pricing) ## Prompt Engineering vs Context Engineering (https://www.slashspace.ai/blog/prompt-engineering-vs-context-engineerings) Published: 2025-12-12 Context engineering and prompt engineering are both related concepts, but they’re different. They both let us engineer an input for the model so that it can do its thing and give us a response. Usually people understand prompt engineering but ignore context. This post tries to unpack the differences between them. If you're new to context or prompts, please read Understanding Tokens and Context first. This post assumes you already know that everything you include has a cost in tokens. Context engineering and prompt engineering are both related concepts, but they’re different. They both let us engineer an input for the model so that it can do its thing and give us a response. Usually people understand prompt engineering but ignore context. This post tries to unpack the differences between them. ## Prompt engineering Prompt engineering is about steering Roles, rules, output format, tone or tool usage policy. A prompt is an instruction. Some examples - Role: “You are a sales assistant for x” - Rules: Never ask for passwords. If unsure, say so - Output Format: Return JSON with these fields. - Tone: Concise, Friendly, technical, etc ## Context Engineering Context engineering is about feeding the model with required information along with the prompt. It could be the history of the conversation, or adding files, media, etc. In a very basic chatbot, the history is already included as part of the context everytime you prompt the model to do something . The single thread of back and forth between you and the model is always sent to the model every time you ask a question. Slightly more advanced chatbots like ChatGPT do additional context engineering such as adding relevant bits of text from other conversations based on your query, get information from a website, etc. ### Who does the prompt engineering and context engineering? As AI users, we most often are doing a bit of both. When you construct a question, you’re engineering the prompt. When you add a file, or reference image, you’re engineering the context. All the AI apps, whether it’s a customer support chatbot, ChatGPT app, Cursor, Lovable, or even RabbitHoles AI, the apps do context and prompt enhancements on top of your prompt and context. This mostly happens behind the scenes. The important thing to know is that the degree of control over context and prompt engineering varies across different kinds of apps built for different usecases. This is where we need to understand the difference between a linear chat versus non-linear chat ## TL;DR * Prompt engineering = how you ask (roles, rules, tone, format). It’s the instruction layer. * Context engineering = what the model knows while answering (chat history, files, retrieved info). It’s the information layer. * Both users and apps do these—apps often add hidden instructions and extra context behind the scenes. ## RabbitHoles AI is now Slashspace AI (https://www.slashspace.ai/blog/rabbitholes-is-now-slashspace) Published: 2026-05-19 This is a bittersweet moment: RabbitHoles AI is becoming Slashspace AI. Many of you found us through the RabbitHoles name, and for a long time, it captured exactly what we were building. But it's time to change as we evolve along with AI market. We're rebranding to Slashspace AI. This is a bittersweet moment: RabbitHoles AI is becoming Slashspace AI. Many of you found us through the RabbitHoles name, and for a long time, it captured exactly what we were building. But it's time to change as we evolve along with AI market. We're rebranding to Slashspace AI. When I first launched, the idea was to have branching conversations inorder to go into rabbitholes about topics that I wanted to learn about. It resonated with many of you. A year and a half later, the AI landscape has changed. Models have become more capable, users have become more sophisticated, and branching conversations alone no longer capture the full opportunity. The name RabbitHoles always carried a sense of curiosity and exploration. But it also suggested something optional; a side quest. What we're building now is meant to be the main quest: the place where AI-native users do their most important work.The product is a UX challenger, challenging the conventional UX of chat window interactions with AI. When I pitched RabbitHoles to new people, most think it's a nice to have tool that can be used someday when they need something. But this doesn't capture the full vision. What we are building is not a nice-to-have research tool. It is a core workstation for AI-native users. Independent knowledge workers like solopreneurs, and researchers. In this direction, we're evolving to implement more agentic features and making the canvas first experience further more intelligent. This means a lot of tweaks to the core UX, and also new features. The name comes from two ideas: Slash: inspired by slash commands, fast actions, and direct interaction.Space: the infinite canvas where your thinking, agents, research, and workflows can live. Slashspace becomes your deep work space. Every day you sit down for deep work, this becomes your default app. Your primary work bench. We're also renaming canvases to spaces. These spaces become your context spaces that can be consumed by other agents as well. More on this coming soon. Over the next few weeks, we'll be launching new features that move us closer to this vision. Thank you to everyone who believed in RabbitHoles. Now, welcome to Slashspace. ## Slashspace 4.0: Quick Window, Stacks and Replays (https://www.slashspace.ai/blog/slashspace-4-0) Published: 2026-10-15 Slashspace 4.0 brings the canvas to wherever you are working. Press Option+/ in any app to ask a question or save a note straight into a canvas, fold piles of nodes into Stacks that keep big canvases fast, and share a canvas as a Replay anyone can watch. Press Option+/ in any app, ask your question, and the answer lands on your canvas. That's the headline of Slashspace 4.0, and it's one of three new features. **Quick Window** lets you ask or jot from anywhere on your Mac. **Stacks** fold a crowded canvas into tidy decks that keep it fast. **Replays** let you share how a canvas came together. Below is what each one does, and the reason it exists.