create_agent

create_agent() builds a LangGraph-backed loop: call the model, run requested tools, return their results, and repeat until the model gives a final response or a stop condition is reached.

Tiny example

from langchain.agents import create_agent

agent = create_agent(
    model="openai:gpt-5.4-mini",
    tools=[get_order_status],
    system_prompt="Help support staff. Never invent order state.",
)

result = agent.invoke({
    "messages": [{"role": "user", "content": "Where is order A-19?"}]
})
print(result["messages"][-1].text)

Agent or fixed chain?

Situation	Start with
The steps are known: validate → fetch → format	Normal Python or a fixed chain
The model must choose among several read-only searches	Agent
A payment, deletion, or external message needs approval	Deterministic policy plus human approval

Failure note

An agent is a loop, so set tool-call limits, timeouts, and cost budgets. Conversation persistence requires a checkpointer and a stable thread ID; create_agent() does not make a stateless model remember previous requests by itself.

Tool calling · Middleware · LangGraph