react-agents-in-langgraph

ReAct Agents in LangGraph

Note: This skill targets LangGraph v1.0+ (released October 2025). The create_react_agent function from langgraph.prebuilt is deprecated in favor of langchain.agents.create_agent. However, create_react_agent remains functional for existing code. For new projects, use from langchain.agents import create_agent (requires LangChain >= 1.0).

ReAct (Reasoning + Acting) is a pattern where LLMs iteratively reason about tasks, select actions (tools), observe results, and continue until reaching a final answer.

ReAct Loop Pattern

Thought -> Action -> Observation -> Repeat until Final Answer

Quick Start with Prebuilt Agent

For rapid prototyping, use the prebuilt create_react_agent:

from langgraph.prebuilt import create_react_agent
from langchain_anthropic import ChatAnthropic
from langchain_core.tools import tool

@tool
def search_tool(query: str) -> str:
    """Search the web for information."""
    return f"Results for: {query}"

# Create model instance (recommended over string format)
model = ChatAnthropic(model="claude-sonnet-4-5-20250929")

# Create a ReAct agent
# DEPRECATED in LangGraph v1.0: Migrate to `from langchain.agents import create_agent`
# For new projects, use: agent = create_agent(model=model, tools=[search_tool], system_prompt="You are a helpful assistant.")
agent = create_react_agent(
    model=model,
    tools=[search_tool],
    prompt="You are a helpful assistant."  # Also accepts SystemMessage, Callable, or Runnable
)

# Run the agent
result = agent.invoke({"messages": [("user", "Search for LangGraph docs")]})

Model Options: Pass a model instance (ChatAnthropic, ChatOpenAI) or use string format "provider:model-name" (e.g., "openai:gpt-4o").

Prompt Flexibility: The prompt parameter accepts: string, SystemMessage, Callable[[list[BaseMessage]], list[BaseMessage]], or Runnable.

Custom Implementation (From Scratch)

For full control over the agent loop, build a custom ReAct graph: