Description
LangWatch is a platform for testing, evaluating and observing AI agents and LLM applications. It runs simulations of realistic user scenarios against an agent, including red-teaming for safety and security gaps, and it lets you write scenarios from Claude Code and run them locally or in CI. Evaluations include LLM-as-a-judge scoring, pairwise comparisons, multimodal checks and online evaluations, and you can search production history with on-demand evals. Observability is OpenTelemetry native, tracing each agent step with cost and latency. The platform also has an AI gateway that sends every AI call through one endpoint with your own keys and budgets per person, team or customer, prompt management with versions, A/B tests and GitHub sync, and a voice AI testing tool. Governance and enterprise controls are included. The project is open source, with code on GitHub, and you can self-host it in about 15 minutes. The pricing is free to start and scales with usage.…
Preview
Highlights
- Agent simulations and red-teaming
- LLM evaluations and judges
- OpenTelemetry-native tracing
- AI gateway with budgets
- Prompt management and self-hosting
Ratings & Reviews
No ratings yet
Be the first to rate LangWatch.
People also visit
OpenLIT: Zero code Agent Observability
Continuous feedback for testing, tracing & fixing AI agents
LangWatch FAQ
What is LangWatch?
A platform for testing, evaluating and observing AI agents and LLM calls, from simulations to production tracing.
Is it free?
The site says it is free to start, and pricing scales by usage.
Can I self-host it?
Yes. The site offers self-hosting, which it says takes about 15 minutes, and the code is on GitHub.
Does it support OpenTelemetry?
Yes. The site describes its observability as OpenTelemetry native.
What is the AI gateway?
A single endpoint for AI calls with your own keys and budgets per person, team or customer.
What are the best LangWatch alternatives?
Popular alternatives to LangWatch include Langfuse, Litefuse and Agenta. See all LangWatch alternatives







