Context Link

Context Link connects nodes on the canvas so an agent can see what another node knows. It works by drawing an edge between two nodes — the same edges you’d use to group work visually — and nodeterm turns that edge into a one-way or two-way context channel.

Linking two agents

Draw an edge between two built-in agent nodes and each side can read the other’s context on demand. It works across all four built-ins — Claude Code, Codex, Gemini, and opencode — in any mix, so a Claude node and a Codex node can read each other.

The read is pull, not push: neither agent is force-fed the other’s history. Instead each agent gets a small helper it can call when it decides the other node’s context is relevant, and that helper prints the linked node’s transcript, a summary, or its recent terminal output. How the helper shows up depends on the agent:

Claude gets a skill it can invoke.
Codex, Gemini and opencode get a short note added to their instructions file (AGENTS.md for Codex and opencode, GEMINI.md for Gemini) describing the command to run.

When you connect the nodes, each session also gets a one-line pointer — dropped in only when the session is idle — so the agent knows the link exists and how to read it.

Linking a note

You can also connect a sticky note to any terminal node. This link is one-way, from the note to the terminal.

When you connect a note to an agent session, the note’s text is pushed into the session once — dropped in when the session is idle so it doesn’t interrupt anything. From then on the note’s live text stays readable through the link: if you edit the note, the agent reads the current text the next time it looks, rather than being interrupted with an update.

A note connected to a plain terminal gets no push — sending text into a bare shell would try to run it as a command. The link still holds the note visually; it just won’t type into the terminal.

See Node types for more on sticky notes and the other kinds of node on the canvas.