Choosing a model

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). Orchestrator mode hides it, because it runs on a model Slashspace manages.

Where AI can run

Option	Setup	What you pay	Where your messages go
Slashspace (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: 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: OpenAI, Anthropic, Google, OpenRouter and more	Paste a key	The provider, per use	Straight from your computer to the provider
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	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 for prices and Privacy and your data 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).

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). 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.
• 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.