What it does
LangWatch collects traces from LLM calls, tool use, and user interactions, then gives teams dashboards, alerts, and online evaluations for production traffic. The platform also includes agent simulations, prompt management, and an AI gateway, but its monitoring view is useful on its own: it helps developers spot a costly or degraded workflow and inspect the trace behind it. Teams can use LangWatch Cloud or run the self-managed version.
LangWatch's monitoring workflow starts by connecting an LLM application through a supported integration or SDK so calls, tool usage, and user events arrive as traces. The observability area exposes trace details and metadata, while dashboards, alerts, and online evaluations turn the incoming data into signals that can be watched over time. When a live evaluation or alert points to a regression, the team can inspect the trace and then use the separate testing and prompt features to verify a proposed change.
The product is broader than a log viewer: it also supports simulated agent testing, evaluation experiments, prompt management, and an AI gateway. Those parts are optional for a team that only needs production observability. LangWatch Cloud's Developer tier includes a monthly event allowance and limited data access; paid tiers add seats, event volume, and retention. The project also documents self-hosting, which trades account-managed service for operating the platform yourself.
The repository describes an Apache 2.0 core with an Enterprise extension, so the open-source claim applies to that core rather than every hosted or enterprise capability. The service detects and helps investigate behavior; a developer must still judge the trace and change the application.
Strengths
- Observability, alerting, and online evaluation use the same production traces, reducing the gap between detection and investigation.
- The docs describe both managed Cloud and self-hosted deployment options.
- The official repository identifies an Apache 2.0 core and makes its scope distinguishable from the Enterprise extension.
Watch out for
- Cloud event volume and retention are plan-limited, so teams with high traffic need to account for usage and storage terms.
- Self-hosting requires a supported Node.js runtime and infrastructure that the team operates.
Best for
- Teams monitoring production LLM and agent traces with custom evaluations or drift alerts.
- Developers who want to turn production failures into reproducible agent tests.
Sources
- Agent Observability & Evals for AI Platform Teams (Oct 26, 2026)
- The open-source platform for AI in production - LangWatch (Oct 26, 2026)
- LangWatch pricing: free to start, scale by usage (Oct 26, 2026)
- langwatch/langwatch (Oct 26, 2026)
- langwatch/LICENSE.md at main · langwatch/langwatch · GitHub (Oct 26, 2026)