First Python agent

Start with a graph that does not call an external model. This verifies your AgentFlow install and the current import paths before you add provider credentials.

This page runs the graph directly in Python:

Create hello_agent.py:

from agentflow.core.graph import StateGraph
from agentflow.core.state import AgentState, Message
from agentflow.utils import END


def assistant(state: AgentState) -> Message:
    latest = state.context[-1].text()
    return Message.text_message(
        f"AgentFlow received: {latest}",
        role="assistant",
    )


graph = StateGraph(AgentState)
graph.add_node("assistant", assistant)
graph.set_entry_point("assistant")
graph.add_edge("assistant", END)

app = graph.compile()

result = app.invoke(
    {"messages": [Message.text_message("Ship the golden path docs.")]},
    config={"thread_id": "get-started-demo"},
)

print(result["messages"][-1].text())

Run it:

python hello_agent.py

Expected output:

AgentFlow received: Ship the golden path docs.

What happened

The graph used the default AgentState, added one node, marked that node as the entry point, and ended after the node returned an assistant message.

The important current imports are:

from agentflow.core.graph import StateGraph
from agentflow.core.state import AgentState, Message
from agentflow.utils import END

Keep the config={"thread_id": "get-started-demo"} value in the example. A thread ID lets the runtime group messages and state for a run, which becomes more important when you add checkpointing later.

Next step

Expose the same graph with the API in Expose with API.