Davia supports LangGraph-style AI agents through the @davia_app.graph decorator.

A graph represents an AI workflow — a sequence of steps connected through logic and memory (called state). It’s useful for building advanced agents like chatbots, assistants, or tool-using flows.

Davia connects directly to the graph you define and builds a UI so users can interact with it — with real-time inputs, state handling, and streaming output.

​
1. What is a graph?

In Davia, a graph is:

  • A LangGraph object (StateGraph) with defined nodes and transitions
  • Exposed to Davia via @davia_app.graph
  • Fully interactive in the frontend — users can provide input, trigger execution, and see the result live

If your agent is conversational, Davia can generate a chat-style interface to let users interact with it naturally.

​
2. Full example: A minimal graph

Here’s a full example of a simple one-turn chatbot using LangGraph and Davia.

from davia import Davia, run_server
from langgraph.graph import StateGraph, START, END, MessagesState

# Define a simple node
def start(state: MessagesState) -> MessagesState:
    return {
        "messages": [
            {"role": "ai", "content": "Hello, how can I help you today?"}
        ]
    }

# Create and expose the graph
davia_app = Davia()

@davia_app.graph
def hello_graph():
    """
    A minimal LangGraph agent that returns a greeting.
    """
    graph = StateGraph(ChatState)
    graph.add_node("start", start)
    graph.add_edge(START, "start")
    graph.add_edge("start", END)
    return graph

#  Launch the server
if __name__ == "__main__":
    run_server(davia_app)

When you run this file, Davia opens the editor at:

🔗 https://dev.davia.ai/dashboard?entrypoint=http://127.0.0.1:2025

You’ll see a simple interface where:

  • Users can submit a message
  • The graph runs and returns a greeting
  • Responses are streamed back to the UI in real time

​
3. How Davia integrates with LangGraph

When you expose a LangGraph agent:

  • Davia reads the state class (e.g. MessagesState) and builds an entire UI based on it,
  • It connects to the graph’s lifecycle: START → node(s) → END
  • It can stream the output live as each node in the graph runs
  • The entire UI is automatically adapted to your graph’s logic

​
4. Best practices

  • ✅ Always return a Graph from your decorated function
  • ✅ Add a helpful docstring to describe your agent
  • ❌ Don’t return the graph directly at the top level — it must be wrapped in a function