LangGraph

LangGraphAgent wraps a compiled LangGraph graph so it can be served through AgentOS or used standalone.

Save the following as langgraph_agent.py. After the Install step, start it with python langgraph_agent.py. Later snippets are alternatives or fragments using the same imports.

from agno.agents.langgraph import LangGraphAgent
from agno.db.sqlite import SqliteDb
from agno.os import AgentOS
from langchain_openai import ChatOpenAI
from langgraph.graph import MessagesState, StateGraph


def chatbot(state: MessagesState):
    return {"messages": [ChatOpenAI(model="gpt-5.4").invoke(state["messages"])]}


graph = StateGraph(MessagesState)
graph.add_node("chatbot", chatbot)
graph.set_entry_point("chatbot")
compiled = graph.compile()

agent = LangGraphAgent(
    name="LangGraph Chatbot",
    graph=compiled,
)

agent_os = AgentOS(
    agents=[agent],
    tracing=True,
    db=SqliteDb(db_file="tmp/agentos.db"),
)
app = agent_os.get_app()

if __name__ == "__main__":
    agent_os.serve(app="langgraph_agent:app", reload=True)

Install

uv pip install "agno[os]" langgraph langchain-openai
export OPENAI_API_KEY=sk-...

Parameters

Parameter	Type	Default	Description
`name`	`str`	`None`	Display name for the agent.
`id`	`str`	`None`	Unique identifier. Auto-generated from `name` if unset.
`graph`	`CompiledGraph`	`None`	Compiled LangGraph graph (`graph.compile()`).
`input_key`	`str`	`"messages"`	Key in the graph state used for input messages.
`output_key`	`str`	`"messages"`	Key in the graph state used for output messages.
`config`	`Dict[str, Any]`	`None`	Optional LangGraph config dict passed to `invoke` / `stream`.
`db`	`BaseDb`	`None`	Database for session persistence.