Laminar

A self-hosted AI agent observability platform with tracing, failure detection and regression evals, available under Apache 2.0 or as a managed service.

Laminar helps developers inspect AI agent runs, find recurring failures and test fixes. It connects production traces with debugging and evaluation tools.

Agent traces: Inspect LLM calls, tool calls, sub-agents, token usage and costs in a transcript, with browser session recordings for browser agents.
Failure detection: Signals analyze runs for unexpected failures and group similar cases; Slack alerts report failures and clusters, and you can ask about traces there.
Regression evals: Turn error clusters into evaluation datasets, run checks locally or in CI, and compare results in the interface.
Coding assistant access: MCP and CLI access let coding agents query trace data with SQL, investigate problems, build evals and verify fixes.
Search and datasets: Search span content, build SQL dashboards and label data from traces or query results for evaluations and fine-tuning.

The open-source platform uses Apache 2.0 and runs on your infrastructure with Docker or Kubernetes, including AWS and GCP deployments. A separate managed service runs on Laminar's infrastructure. Its OpenTelemetry tracing integrates with Claude Agent SDK, OpenAI Agents SDK, Vercel AI SDK, LangChain and Browser Use.

AI-assisted trace chat, SQL generation and server-side AI workers require an LLM provider. Self-hosted deployments collect anonymized usage telemetry, which you can disable.