ReAct: The Core Pattern

The Reason-Act loop that powers every LangGraph agent: think, pick a tool, observe the result, repeat until done.

Quick Reference

→ReAct = Reasoning + Acting: the LLM alternates between generating thoughts and executing tool calls
→LangGraph implements ReAct as a cycle: agent node (LLM) → tool node → agent node → ... → END
→The agent decides to stop when it generates a response without any tool_calls in the AIMessage
→Structured tool schemas with clear descriptions are critical — the LLM reads them to decide which tool to use
→Add a recursion_limit to the graph to prevent infinite loops (default is 25 in LangGraph)

What is ReAct?

Definition

ReAct (Reason + Act) is the foundational agent pattern: the LLM generates a thought about what to do next, executes a tool action, observes the result, and repeats until it has enough information to respond.

Every agent framework implements some variation of ReAct. In LangGraph, it maps directly to a graph with two nodes and a conditional edge: the agent node calls the LLM, the tool node executes tool calls, and a conditional edge checks whether the LLM is done or wants to call more tools.

The loop runs until the LLM decides it has enough information to respond

ReAct in LangGraph

The complete ReAct agent in LangGraph — 20 lines

Thought → Action → Observation

Step	Who	What happens	LangGraph equivalent
Thought	LLM	Reasons about what to do next	agent node — model.invoke()
Action	LLM	Generates tool_calls in AIMessage	Part of agent node output
Observation	System	Executes tools, returns results	tools node — ToolNode()
Repeat	System	Checks if agent is done	tools_condition edge
Response	LLM	Generates final answer (no tool_calls)	agent node → END

Sign in to read this article

This is a premium article. Sign in with your Google account to continue.