QuickstartMoving from RabbitholesCore concepts
Canvas OverviewCreating NodesConnectionsOrganizing the CanvasSearchSide ViewExportKeyboard Shortcuts
Chat NodeText NodeWeb NodePost Node (YouTube & more)Document Node (Beta)Image NodeGroup Node
Models & ProvidersRAG ModePersonasSlash Commands (Skills & Prompts)Using MCP Tools in ChatSub-AgentsVoice Input
Cloud ProvidersAgent ProvidersOllama (Local Models)Custom ProvidersMCP Servers (Beta)Default ModelsConnected ToolsPersona PromptsAccount ManagementSystem Preferences
Troubleshooting & Support

Connections

Connect nodes to control exactly what context the AI sees — handles, branches, context modes, and token summaries.

Last updated: July 6, 2026

Connections are what make the canvas more than a whiteboard. Every edge you draw is a statement about context: "the node on the right can see the node on the left." Instead of stuffing everything into one long chat, you engineer the exact context each conversation gets — which means more relevant answers, fewer tokens, and no context pollution between unrelated threads.

Handles: givers and takers

Every node has two connection points:

  • Right handle (giver) — the source. It turns orange when connected. Drag from here to send this node's content onward.
  • Left handle (taker) — the target. Hover it to see a tooltip with the incoming context size, e.g. "1,204 incoming tokens".

To connect two nodes, drag from one node's right handle to another node's left handle.

You can't connect a node to itself ("Cannot connect a node to itself."), you can't create duplicate edges between the same pair, and Image nodes only accept other Image nodes as input ("Image nodes can only accept images nodes as input").

How context flows

Context flows left to right and chains through the whole upstream path. If A → B → C, then C sees both A and B. This is the core mental model:

  • A chat node reads everything connected upstream of it — text notes, web pages, documents, other chats — as context for its conversation.
  • Adding or removing an edge immediately changes what the AI knows in that chat.

You can also connect nodes from inside a chat: typing @ in a chat node's input lets you mention another node, which creates the edge for you. See Chat Node.

Use cases

  • Research: connect three source Web nodes into one chat node and ask it to synthesize them — the answer draws only on those sources.
  • Writing: chain outline → draft → editor-chat, so the editing conversation sees both the outline and the draft.
  • Studying: connect a lecture PDF (Document node) to several chat nodes, one per exam topic, and quiz yourself in each without the threads bleeding into each other.
  • Planning: keep a "project brief" Text node upstream of every chat on the canvas so all of them share the same ground truth.

Branches

A branch is a new chat node that continues from an existing node:

  1. Double-click the right handle of any node — a tooltip says "Create a new branch" — and a connected Chat node is created.
  2. Or press ⌘B with a node active to branch from it.
  3. Inside a chat, ⌘⇧Enter sends your typed message into a brand-new branch instead of the current thread.

Use cases

  • Mid-conversation, you want to explore a tangent without derailing the thread — branch it, and the tangent inherits all the context up to that point.
  • Compare answers: branch the same source twice and ask each branch with a different framing or model.

Context modes

Chat nodes have a context-mode toggle on their left edge:

  • Summarized context (zap icon) — the default. Incoming context is summarized and passed into the conversation.
  • Isolated context — the chat ignores all incoming connections and runs standalone.

The toggle is disabled while the chat is generating a response.

Use cases

  • Flip a scratch-pad chat to Isolated context so a quick side question doesn't burn tokens re-reading a large upstream document.
  • Keep a "devil's advocate" chat isolated so it critiques your conclusion without being anchored by the sources.

Token summary

Every chat node shows a small token pill (the asterisk icon). Click it to open the Token Summary popover, which breaks down what will be sent to the model:

  • Incoming — tokens from connected upstream nodes (shows 0 in Isolated mode).
  • Own — tokens from this node's own messages.
  • Input — tokens in the message you're currently typing.
  • Total — the sum.

Treat these numbers as an estimate — they're tokenized with the OpenAI tokenizer, and other providers count slightly differently.

Use cases

  • Before sending an expensive request, check whether an upstream document is dumping 50k tokens into the prompt — and disconnect or isolate if so.
  • Debug a "context too long" error by finding which upstream branch carries the weight.

Edge styles and editing

  • Choose from five edge styles in canvas settings: Default, Straight, Step, Smooth step, or Simple bezier.
  • Select an edge and press Backspace/Delete to remove it.
  • Drag either end of an existing edge to reconnect it to a different node.

Creating Nodes

Every way to add nodes to the canvas — toolbar, right-click, Cmd+click, drag-and-drop, and paste.

Organizing the Canvas

Select, align, group, color, copy, and auto-arrange nodes to keep large canvases readable.

On this page

Handles: givers and takersHow context flowsBranchesContext modesToken summaryEdge styles and editing