langviz-studio 0.1.1

LangViz-Studio is an observability toolkit for LangGraph workflows.

LangViz-Studio

LangViz-Studio is an observability toolkit for LangGraph workflows.

It enables you to track the execution states, node transitions, and conditional edges of your state-based flows or agent pipelines in real-time, sending all data asynchronously to a customizable endpoint (e.g., a Next.js server) for visualization.

Key Features

• GraphObservability

  • Automatically records and sends node start/end events, edge transitions, and conditional branch usage.
  • Maintains a unique graphId (UUID by default) and an optional friendly graph_name.
  • Sends data to a configured endpoint (http://localhost:3000 by default).

• ObservabilityStateGraph

  • A subclass of langgraph.graph.StateGraph that automatically wraps node functions and edge definitions so all observability data is captured with no extra code.
  • Just use add_node(...), add_edge(...), add_conditional_edges(...) as normal, and your graph’s runtime states will be sent in real-time for visualization.

• Asynchronous and Non-blocking

  • Uses an internal background thread running an asyncio event loop to POST updates via httpx.AsyncClient, ensuring minimal impact on your main workflow thread.

Installation

pip install langviz-studio

(Also ensure you have installed langgraph and other dependencies like langchain_core, if applicable.)

Usage

Below is a minimal example showing how to use LangViz-Studio with LangGraph. In this example, we define two simple asynchronous node functions (run_start and run_end) and connect them in a StateGraph. Observability logs are sent automatically.

import asyncio
from langviz_studio.observability import GraphObservability, ObservabilityStateGraph

async def run_start(state, config=None):
    print("=== [start] node invoked ===")
    state["msg"] = "Hello from start!"
    return state

async def run_end(state, config=None):
    print("=== [end] node invoked ===")
    state["result"] = "Done"
    return state

async def main():
    # 1. Set your Observability
    obs = GraphObservability(
        graph_name="MyExampleGraph"
    )

    # 2. Using ObservabilityStateGraph instead of StateGraph
    workflow = ObservabilityStateGraph(obs, state_type=dict)

    # 3. Nodes are added in the same manner as StateGraph
    workflow.add_node("start", run_start)
    workflow.add_node("end", run_end)
    # Edges / Conditional edges are also added in the same manner as StateGraph
    workflow.add_edge("start", "end")

    workflow.set_entry_point("start")

    # 4) Compile the graph
    compiled = workflow.compile()

    initial_state = {}
    # 5) Run the workflow
    result = await compiled.ainvoke(initial_state)
    print("=== Workflow finished ===")
    print("Final State:", result)

if __name__ == "__main__":
    asyncio.run(main())

Server-Side Visualization

By default, GraphObservability calls two endpoints on your server:

• POST /api/graph/start — triggered once when the first node starts, to initialize a new graph record.
• POST /api/graph/ingest — triggered after every node start/end or edge creation, sending updates.

How it Works Internally

1. A background thread holds an asyncio event loop (via _AsyncLoopThread), so we can await httpx.AsyncClient.post(...) without blocking the main process or requiring the user to manage async.

2. Each node start/end or edge creation calls a short, synchronous method (_post_to_nextjs) which enqueues a coroutine in the background event loop to do an HTTP POST.

3. This ensures minimal overhead and “fire-and-forget” updates to your server.

License

MIT License. See LICENSE for details.

Contributing

Pull requests and issues are welcome! If you have suggestions or find bugs, please open an issue or submit a PR on the GitHub repository.