Skip to main content

React Agent

create_agent builds a ReAct-style agent — reason, act, observe, repeat — from a model and a tool list. The constructor itself lives in LangGraph's prebuilt layer; langchain.agents.create_agent is the stable entry point.

from langchain.agents import create_agent

def search_docs(query: str) -> str:
    """Search internal documentation."""
    return f"Top result for '{query}': see the onboarding guide, page 4."

def get_ticket_status(ticket_id: str) -> str:
    """Look up the status of a support ticket."""
    return f"Ticket {ticket_id}: in progress, assigned to support."

agent = create_agent(
    model="claude-sonnet-4-6",
    tools=[search_docs, get_ticket_status],
    system_prompt="You are a support assistant. Use tools to answer accurately.",
)

result = agent.invoke({
    "messages": [{"role": "user", "content": "What's the status of ticket 4821, and where's the onboarding guide?"}]
})
for msg in result["messages"]:
    print(type(msg).__name__, "-", getattr(msg, "content", msg))

Walking one trace through the loop:

  1. Reason — the model reads the question, decides it needs two tool calls (one per sub-question).
  2. Act — it emits an AIMessage with two tool_calls: get_ticket_status and search_docs.
  3. Observe — create_agent executes both, appends a ToolMessage per call.
  4. Reason again — the model reads both results and decides it has enough to answer.
  5. Act (final) — it emits an AIMessage with no tool_calls, just text. The loop stops here.

See also​