Lesson 15 / 25

The AI Agent Node

Assemble an agent from a chat model, memory and tools and limit how long it may run.

A node with three plug-ins

The AI Agent node runs a tool-calling loop for you. Attach a chat model (required), optional memory so a chat keeps its history, and one or more tools it may call. Write a clear system message describing the role, the tools and the rules. Set a maximum number of iterations so a confused agent cannot loop forever.

A system message for an agent

Say which tool to use when, what to refuse, and how to answer. Short and specific beats long and vague.

You are a support assistant for Acme.
- To check an order, call get_order with the order number.
- To answer policy questions, call search_policy first and quote it.
- Never issue refunds yourself; create a refund request with request_refund.
- If you cannot find the answer, say so and offer to contact a human.
- Reply in the customer's language, in under 120 words.

Cap the iterations

Keep the maximum iterations low (for example 5 to 10) for simple agents. Each iteration is another model call, so unbounded loops directly cost money and time.

Quick check: Which sub-node is required to run an AI Agent?

  • A chat model
  • A vector store
  • A Slack node
  • A Wait node
Answer

A chat model — The model provides the reasoning; memory and tools are optional extras.