langwatch/langwatch

Open-Source Agent Observability & Evals for AI Platform teams. Trace, test, route and govern every agent & coding assistant in the company.

LangWatch – Observability & Governance for LLM‑powered Apps

LangWatch is an open‑source platform that helps teams run, monitor, test, and govern every call to large language models (LLMs) inside a company. It works with coding assistants (Claude Code, GitHub Copilot, etc.), LLM agents built on LangChain/LangGraph, and any OpenAI‑ or Anthropic‑compatible endpoint. The core is Apache‑2.0; an optional Enterprise extension adds extra features and requires a commercial licence.

What the platform does

Area	What you get
LLM Ops	Real‑time observability of LLM calls, prompt versioning, simulation‑based testing, and evaluation dashboards.
Coding Agents	Track usage of coding assistants per pull‑request or team, see cost breakdowns, and enforce privacy controls.
AI Gateway	A single OpenAI/Anthropic‑compatible endpoint that can issue virtual keys with budget limits, route requests to different models, and apply rate limits.
AI Governance	Inventory of every AI tool in the org, usage reports, and anomaly‑detection rules (e.g., unused Copilot subscriptions).

Quick start

Step	Command / Action
Try the hosted cloud	Visit https://app.langwatch.ai and sign up.
Track a coding agent locally	`npx langwatch claude` (replace `claude` with `codex`, `copilot`, `opencode`, …)
Self‑host the server	`npx @langwatch/server` – only Node.js is required.

The repo also publishes a Python SDK (pip install langwatch) and an npm package (npm i langwatch). Detailed installation steps are in the docs.

Integrations (as listed in the README)

Coding assistants – Claude Code, OpenAI Codex, GitHub Copilot CLI, Opencode, Cursor, MCP server, etc.
Frameworks – LangChain, LangGraph, Vercel AI SDK, Mastra, CrewAI, Google AI SDK, DSPy, OpenTelemetry.
Model providers – OpenAI, Anthropic, Azure OpenAI, Vertex AI, AWS Bedrock, plus others.
Governance sources – Copilot Studio, Claude Cowork, compliance APIs from Anthropic/OpenAI, Workato, Databricks Genie, OpenTelemetry/S3 audit feeds.
No‑code platforms – LangFlow, Flowise, n8n.

Typical use cases

Simulation testing – Write test scenarios that replay LLM calls to benchmark agents or voice assistants.
Cost optimisation – Measure cost per PR for Claude Code and adjust prompts to stretch usage limits.
Budgeted access – Issue virtual API keys with per‑key spending caps for internal teams or external customers.
License hygiene – Identify under‑used AI subscriptions (e.g., Copilot Studio) and cancel them.

Licensing & security

Core platform: Apache 2.0.
Enterprise modules (in enterprise/): commercial licence required for production use.
SDKs are MIT‑licensed.
Claims GDPR compliance and ISO 27001 certification; security issues can be reported to security@langwatch.ai.

Where to learn more

Docs & tutorials – https://langwatch.ai/docs/introduction
Discord community – https://discord.gg/kT4PhDS2gH
Contributing – see CONTRIBUTING.md and run make quickstart for a local dev environment.

Bottom line: LangWatch is a production‑ready observability, testing, and governance stack for any organization that relies heavily on LLMs and AI agents, offering both a free open‑core version and a commercial enterprise edition.