Skip to main content

Steps — Basics

Intermediate steps in a conversation-mode Tool behave identically to steps in a regular Tool. All node types are available, all configuration options work the same way, and runtime behavior is the same.

For the complete step authoring guide, see Tool steps — Basics.

What's the same as a regular Tool​

  • The full step library is available — LLM, Knowledge Base Search, API, Python, Web Research, Web Scraper, Google Search, LVM, PDF/Image/Audio utilities, Translate, Delay, MCP, and more.
  • Variable referencing with {{ }} Jinja templating.
  • Execution Rules — Trigger rule, Condition, Foreach.
  • Guardrails on a per-step basis.
  • Run a step individually or run the full chain.
  • Adding, deleting, and reordering steps.

What's different in Conversation mode​

Only two things differ — both at the edges of the graph, not in the middle:

  • Start Point input is Chat History, not typed inputs. See Start Point: Chat History.
  • Final node is Final Output with a single template, not the multi-key Outputs node. See Final Output.

Every intermediate step between these two endpoints is configured exactly as you would in a task-mode Tool.

Where to learn step authoring​

Reminder for chat agents: In a conversation-mode Tool, the value you map in the Final Output node becomes the Tool's reply to the user. Plan your step graph so the variable you want returned is available there.