LangGraph Observer

LangGraph Observer is an observability dashboard designed to enhance the visualization and interaction with LangGraph workflows. This project offers a compact environment that integrates a FastAPI backend and a Streamlit dashboard, facilitating the efficient running and inspection of large language model (LLM) operations.

Key Features

• FastAPI Backend: Executes the underlying graph and provides robust API endpoints for workflow execution and health checks.
• Interactive Streamlit Dashboard: Users can interactively prompt the model, view metrics, and manage the workflow outputs seamlessly.
• Modular Services: Includes functionalities for generation, emoji transformations, toxicity assessment, hallucination scoring, and artifact logging.
• Artifact Management: Each workflow run generates artifacts that document output, scores, token usage, duration, and cost, enabling comprehensive traceability.

Workflow Pipeline

The typical workflow consists of the following steps:

1. Generate output from the LLM.
2. Optionally transform the output into an "emoji-fy" format.
3. Calculate the emoji-ness score.
4. Assess toxicity metrics.
5. Evaluate for hallucination tendencies.
6. Save generated artifacts and append run history.

Dashboard Overview

The dashboard comprises various interactive features:

• Input prompts and view sample prompts.
• Run Workflow button to execute the pipeline.
• Make More Emoji button for iterative transformations.
• A sidebar displaying essential metrics like toxicity, hallucination scores, and execution duration.
• A table to review recent runs along with full-state JSON outputs.

API Interface

The FastAPI backend exposes two main endpoints:

• POST /run-graph: Triggers the complete workflow.
• GET /health: Provides a health status check for the API.

Getting Started

To launch the Streamlit dashboard, use the command:

uv run streamlit run app/dashboard/ui.py

To start the FastAPI server, run:

uvicorn app.api.server:app --reload

Project Structure

The project is organized as follows:

app/
  api/
  adapters/
  dashboard/
  domain/
  services/

LangGraph Observer exemplifies an advanced solution for monitoring and interacting with LLM workflows, combining ease of use with powerful observability features.