laminar

Laminar is an open-source observability platform built specifically for AI agents. It monitors every agent run, detects failures, and helps you understand what went wrong — all the way from a high-level alert to the exact step that caused the issue. Think of it as a debugger and monitoring dashboard rolled into one, designed for the messy, multi-step nature of modern AI agents.

What makes Laminar stand out is how it turns raw agent traces into actionable information. You set up Signals — plain-English rules like "agent is stuck in a loop" — and Laminar watches every run, pinging you on Slack when it matches. The trace viewer lays out inputs, LLM reasoning, tool calls, and sub-agents in a clear timeline. You can even ask natural language questions about a run and get answers with direct links to the relevant context. Events are automatically grouped into clusters — "Tool call failures", "LLM behavior issues" — and Laminar tracks them over time, resolving or reopening clusters automatically.

Laminar is built for developers building and maintaining AI agents in production. If you're using frameworks like Vercel AI SDK, Claude Agent SDK, or OpenAI Agents SDK, integration takes two lines of code. The CLI and MCP support let coding agents like Claude fully own the debugging loop: run the agent, read the trace, fix, and re-run with cached state. Teams also use it to turn fixed issues into eval datasets, catching regressions before they ship. Self-hostable, HIPAA and SOC 2 compliant, with a free tier to get started.