create_agent: Reference Guide

What is create_agent?

create_agent() is the high-level entry point in LangChain v1 for building tool-calling agents. It replaces the older initialize_agent and AgentExecutor APIs with a single function that wires up a ReAct-style loop for you: the model is called, any tool calls it emits are executed, the results are fed back, and the cycle repeats until the model returns a final answer. Under the hood it builds and compiles a LangGraph graph, so what you get back is a fully-featured Runnable with invoke, ainvoke, stream, and astream.

You pass a chat model and a list of tools; optionally a system prompt to steer behaviour, middleware to hook into each step, a checkpointer to persist conversation state, and a response_format for structured final output. Because the result is a graph, it speaks the messages-based state convention: you invoke it with {"messages": [...]} and read the final answer from the last message in the returned state. Tools are bound automatically — you do not call bind_tools yourself.

create_agent is the recommended starting point for most agents because it removes boilerplate while staying inspectable: you can stream intermediate steps, attach a MemorySaver checkpointer for multi-turn memory, and add human-in-the-loop interrupts through middleware. When you need control it cannot express — custom branching, parallel tool fan-out, or bespoke state — drop down to StateGraph directly and build the loop yourself. The two share the same runtime, so graduating from create_agent to a hand-written graph is incremental, not a rewrite.

When to Use

You're building an agent. Use create_agent() for the recommended v1 approach.

Use Cases

• • Create LLM agents
• • Tool-using AI
• • Agentic workflows
• • Auto task solving
• • Autonomous agents
• • Multi-step reasoning

Key Features

• ✓ Simple creation
• ✓ Auto tool binding
• ✓ Middleware support
• ✓ State management
• ✓ Checkpointing
• ✓ Streaming

When NOT to Use

Notes

Import

from langchain import create_agent

Key Parameters

Code Examples

from langchain import create_agent
from langchain_core.messages import HumanMessage
from langchain_openai import ChatOpenAI

model = ChatOpenAI(model='gpt-4o')
agent = create_agent(model, tools=[search_tool, calculator_tool])
result = agent.invoke({'messages': [HumanMessage(content='What is 23 * 17?')]})
print(result['messages'][-1].content)

from langchain import create_agent
from langchain_openai import ChatOpenAI

agent = create_agent(
    ChatOpenAI(model='gpt-4o'),
    tools=[search_tool],
    system_prompt='You are a concise research assistant. Cite sources.',
)
for step in agent.stream({'messages': [HumanMessage(content='Latest on RAG?')]}):
    print(step)

from langchain import create_agent
from langgraph.checkpoint.memory import MemorySaver
from langchain_openai import ChatOpenAI

agent = create_agent(
    ChatOpenAI(model='gpt-4o'),
    tools=[search_tool],
    checkpointer=MemorySaver(),
)
config = {'configurable': {'thread_id': 'user-42'}}
agent.invoke({'messages': [HumanMessage(content='I live in Berlin')]}, config)
agent.invoke({'messages': [HumanMessage(content='What is the weather here?')]}, config)

Common Mistakes

Alternatives

Related LangChain References

Browse the full LangChain API reference index to explore more classes, methods, and decorators, or start with the LangChain introduction tutorial for end-to-end context on building with create_agent and the wider framework.

create_agent FAQ