@langgraph-js/smith-server

@langgraph-js/smith-server is an open-source backend for langgraph-js that offers LangSmith-compatible tracing and observability for your AI applications. It provides a self-hostable alternative, giving developers full control over their tracing data.

Features

• LangSmith Compatibility: Designed to work seamlessly with langgraph-js to capture traces in a LangSmith-like format.
• End-to-End Tracing: Capture and visualize the entire lifecycle of your AI runs, from user input to final output.
• Detailed Run Information: Log inputs, outputs, events, errors, and metadata for each step of your AI application.
• Feedback Mechanism: Collect and store feedback on individual runs to improve your models and applications.
• Attachment Support: Store relevant files and data associated with specific runs.
• System & Thread Tracking: Organize and filter runs by system and conversation thread for better management and analysis.
• Database Agnostic: Supports both SQLite (for lightweight local development) and PostgreSQL (for production-ready deployments) through a flexible adapter pattern.

Installation

pnpm i -D @langgraph-js/smith-server
pnpm open-smith

After installation, configure your LangGraph project to use this server by setting the LANGSMITH_ENDPOINT environment variable:

LANGSMITH_ENDPOINT="http://localhost:7765"

GUI Admin URL:

http://localhost:7765/ui/index.html

Use PG

By default, open-smith uses SQLite to store trace data. You can configure it to use PostgreSQL instead:

To use PostgreSQL, add the following to your .env file:

TRACE_DATABASE_URL=postgresql://postgres:postgres@localhost:5434/open_smith