Global Chat is the always-on assistant that can run any workflow, stream results, and escalate to autonomous agents when a task needs planning. Use this page to jump to the right references.
Core guides
- Global Chat — Interface layout, thread management, and how chat invokes workflows or tools.
- NodeTool Agent System — How planning, tool usage, and execution loops work under the hood.
- Global Chat CLI & Chat Server — Automate conversations or integrate with custom frontends.
- Chat API — Programmatic access for running chats, streaming outputs, and issuing tool calls.
Typical flows
-
Run any saved workflow from chat
Save the workflow in the editor, open Global Chat, and choose Run workflow in the composer—perfect for ad-hoc runs or sharing with non-builders. -
Enable Agent Mode for planning tasks
Switch to Agent Mode to let the LLM break goals into subtasks, call frontend tools (add nodes, align graphs), and update the workflow automatically. See the Agent guide. -
Expose agents via APIs or CLI
Use the Chat CLI for automation scripts, or call the Chat API from backend services so agents can operate in headless environments. -
Stream updates back into editors
Workflow changes generated by agents propagate to open editors via WebSocket updates, keeping the canvas synced without manual refresh.
Ready to go deeper? Pair this with the Cookbook for agent patterns or the Deployment Guide to move agents into production.