Laminar
Open-source agent observability with traces, failure detection and evals
Laminar is an OpenTelemetry-native observability platform for AI agents. It captures LLM calls, tool calls, sub-agents, costs and tokens, and renders each run as a readable transcript rather than raw spans.
Signals works in both directions: describe a behaviour in plain English and it reads every trace for matches, and it also surfaces failure modes that were never defined. Similar events cluster, and a cluster can become an eval dataset.
Apache 2.0, self-hostable with Docker or Helm. Full SQL access over traces, signals and eval scores from the UI, CLI, API or MCP.
Pricing: Monthly subscriptions
What is Laminar?
Laminar is an open-source observability platform built for AI agents rather than single LLM calls. A line of setup instruments an existing stack, traces are sent in the background, and each run is shown as a transcript to read top to bottom.
It is OpenTelemetry-native and Apache 2.0 licensed, self-hostable with Docker or a Helm chart on AWS or GCP.
Signals: finding failures you did not define
Most tools in this category record what happened and leave the reading to you. Signals runs over every trace automatically and works two ways. A behaviour can be described in plain English, such as "the agent looped without progress", and Laminar reads every trace for it and writes a structured event on a match. It also surfaces failure modes that were never specified.
Similar events are grouped into clusters, so a new report can be checked against whether the same failure has happened before and how often. Slack alerts fire when something breaks, so a bad run surfaces without going looking for it.
Debugging and regression testing
The debugger sets a checkpoint on any span and reruns from that point, with the earlier steps served from cache rather than re-executed. Any error cluster can be turned into an eval dataset, run locally or in CI, with runs compared side by side.
Traces, spans, signal events and eval scores are all queryable in raw SQL from the UI, CLI, API or MCP server, so a coding agent can investigate its own failures.
Integrations
Two lines of setup for the Vercel AI SDK, Claude Agent SDK, OpenAI Agents SDK, LangChain, Mastra, Pydantic AI, OpenHands, LiteLLM, Browser Use, Stagehand and Playwright, plus the OpenAI, Anthropic and Gemini SDKs. Browser session recording is synced to the trace for browser agents.
Who it fits
Teams running agents in production where a single run is dozens of steps and there are too many runs to read by hand. The question there is which step broke across thousands of runs, not what one prompt returned. Self-hosting covers data residency requirements.
Laminar Alternatives
Explore 55 products in the Observability & Analytics category. View all Laminar alternatives.
Langfuse
Traces, evals, prompt management and metrics to debug and improve your LLM application.
Klu
Collaborate on prompts, evaluate, and optimize LLM-powered Apps with Klu.
Agenta
Open-source prompt management, evaluation, and observability for LLM apps
AiDren
Drop-in security proxy that blocks prompt injection and data leaks in LLM apps by changing one base URL
Lunary
The platform to monitor, manage and improve your LLM apps.
Also listed in
Work on Laminar? Feature it at the top of Observability & Analytics.
Is your product missing?