★ OverviewIntermediate9 min

create_agent

create_agent builds a graph-based agent runtime on top of LangGraph. Give it a model and tools — it handles the reasoning loop, tool dispatch, and stopping conditions.

Quick Reference

→from langchain.agents import create_agent
→Returns a CompiledStateGraph — supports .invoke(), .stream(), checkpointing
→Agent loop: model reasons → calls tools → observes results → repeats until done
→Stops when model emits final text (end_turn) or max iterations is reached
→Pass string model ID: create_agent('anthropic:claude-sonnet-4-20250514', tools=[...])

What create_agent Builds

create_agent compiles a full agent runtime into a CompiledStateGraph backed by LangGraph. It is not a simple chain or a runnable — it is a stateful graph that alternates between a model node (which calls the LLM) and a tools node (which executes tool calls). Because it returns a CompiledStateGraph, you automatically get .invoke(), .stream(), checkpointing, middleware hooks, and human-in-the-loop support — none of which require additional wiring.

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

Minimal Setup

A working agent in 8 lines

The Agent Loop

The agent follows the ReAct pattern: Reason, then Act. On each iteration, the model sees the full message history and decides whether to call a tool or emit a final answer. If it calls a tool, the result is appended to the message history as a ToolMessage and the loop continues. If it emits text without a tool call, the loop ends and that text is the final response.

Sign in to read this article

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