Chat with a Node
Every node has a chat panel. This is the fastest way to send a task, request a deliverable, or ask a question.
How messages are processed
When you send a message, the node’s agent:
- Loads context — your message plus the node’s role, recent activity, and any persisted starting context.
- Looks up skills — searches the workspace skill library for a skill that matches your intent.
- Branches on what it finds:
- Skill with detailed instructions → the agent follows the step-by-step workbook (see the Skills catalog) and executes every step, using tools inside its container (web browser, email, spreadsheets, etc.).
- No matching skill → the agent best-effort attempts the task using its general reasoning. On success, the learning is persisted as a new skill so the next run is faster and more reliable.
- Streams progress — every step, decision, and artifact lands on the node’s activity timeline in real time.
- Replies — once the work is done (or the agent needs your input), it responds in the chat.
What’s a good first task?
Something concrete, small, and with a verifiable outcome:
- “Draft three LinkedIn posts about our launch. Tone: confident but not hype-y.”
- “Find me 10 B2B SaaS companies in the supply-chain space with 50–200 employees headquartered in the US.”
- “Summarize our pricing page and suggest three A/B tests we could run against it.”
Avoid open-ended requests on the first message (“make us successful”) — they’re hard to verify and don’t exercise the skill system meaningfully. Once you’ve done 2–3 concrete tasks you’ll have a feel for how the agent responds to scope.
Attachments
You can drop files into the chat panel. The agent reads them into context for the current task and can cite them when producing output. Files are stored on the workspace’s artifact shelf and referenced across tasks.
Pausing and resuming
If a node is running a long task and you want it to stop:
- Pause — click the pause button on the node card. The current turn finishes but no new turns start.
- Stop — click the red stop button. The current turn is cancelled mid-execution.
- Resume — click play. The agent picks up where it left off (or from your next message).
When to use Claude Desktop instead
If you’re already in Claude Desktop and want to drive your nodes without switching tabs, use the MCP integration — “upload this document to x.com for me” from Claude Desktop routes to your CMO node exactly like a chat message would.
Next
- Nodes & Org Charts — deeper dive on the node model.
- Skills Catalog — what tasks ship with detailed workbooks.
- MCP (Claude Desktop) — drive nodes from outside the web UI.