Trace agent runs
Capture agent runs as readable traces and timelines that show input, LLM reasoning, tool calls, and sub-agents. This helps teams inspect what happened without stepping through raw logs.
Laminar
ReclamarLaminar is an open-source observability platform for AI agents to trace runs, detect failures, and run evaluations with Slack alerts and self-hosting.
Open-source observability for AI agents
Laminar is an open-source observability platform for AI agents. It is built to trace agent behavior, evaluate changes, and help teams find the exact step where a run went wrong.
The homepage centers the product around three workflows: catching failures with Signals, inspecting runs through a readable transcript and timeline, and turning recurring issues into evaluations. The site also describes developer-oriented capabilities such as a debugger session flow, CLI and MCP access, full-text search, SQL queries, dashboards, and browser session recording.
The pricing page shows free, Hobby, Pro, and Enterprise options, with the public tiers covering trace ingestion, signal usage, retention, support, and platform features. Laminar also lists support for multiple agent SDKs and browser-agent tools, which suggests it is meant to fit into existing AI application stacks.
The site positions Laminar for teams building or operating AI agents that need visibility into failures, repeatable evaluations, and a way to move from issue detection to fixes and regression checks.
Features
Signals and Slack alerts
Describe problems in plain English and let Signals watch for them across runs. Laminar can ping Slack when a run matches the issue, turning recurring failures into alerts.
Issue clustering
Group repeated issues into named clusters and track them over time. When a cluster stops recurring, Laminar resolves it and can reopen it if the issue returns.
Evals from failures
Turn fixed error clusters into evaluation datasets and run evals after changes to catch regressions. The product page shows this workflow for comparing model or agent behavior across releases.
Debugger loop
Use the debugger session flow to run an agent, read the trace, apply fixes, and rerun with cached state. The site also mentions Laminar CLI and MCP for this workflow.
Use cases
Debug production agent failures
Use Laminar when you need to understand why an agent failed in production. The trace view and Signals workflow help isolate the step, tool call, or model behavior that caused the issue.
Monitor recurring run issues
Use the product to keep watch for known failure patterns such as loops, wrong directory usage, or repeated command errors. Signals can detect the issue across runs and send Slack alerts when it appears.
Iterate on agent behavior
Use Laminar when you want a fix-and-verify loop for agent development. The debugger session flow, CLI, and MCP support reruns with cached state so you can inspect and iterate on changes.
Build evaluation datasets from incidents
Use the platform to turn resolved failures into regression checks. Fixed error clusters can become eval datasets, which helps teams compare behavior before and after a release.
Analyze and annotate agent data at scale
Use the search, SQL, dashboards, and browser recording features when your team needs to analyze traces at scale or review browser-agent sessions as part of debugging and annotation.
Pros and Cons
Pros
- Open-source platform with Apache 2.0 licensing and self-hosting options.
- Readable trace views surface LLM reasoning, tool calls, and sub-agents in one place.
- Signals can alert on plain-English failure descriptions and route alerts to Slack.
- Recurring issues can be clustered and converted into evaluation datasets.
- The pricing page clearly separates free, self-serve, and enterprise paths.
Cons
- Some platform capabilities are only mentioned briefly on the pricing page, so the public marketing pages provide limited detail on implementation and setup.
- The source does not show a separate integrations or docs page, so compatibility details are easier to verify for listed SDKs than for the full platform surface.
FAQ
What is Laminar used for?
Laminar is built to trace AI agent runs, surface failures, and run evaluations so teams can debug and improve behavior over time. The homepage also describes Signals for alerting and cluster tracking for recurring issues.
How does Laminar help with debugging?
The product page says Laminar makes agent runs navigable through a transcript and timeline that surface input, LLM reasoning, tool calls, and sub-agents. It also offers a debugger flow that can rerun agents with cached state.
Does Laminar have a free plan?
The pricing page lists a free plan, Hobby, Pro, and Enterprise. Enterprise is custom priced with custom limits and on-premise deployment, while the public plans show different data, signal, retention, and support tiers.
What tools and frameworks does Laminar support?
The site lists integrations and compatibility with Claude Agent SDK, OpenAI Agents SDK, Mastra, Pydantic AI, LangChain, OpenHands SDK, Browser Use, Stagehand, Playwright, Anthropic, OpenAI, and LiteLLM. It also mentions Laminar CLI and MCP access.
What core platform features are included?
The pricing page notes features such as OTLP trace ingestion, full-text trace search, custom dashboards, SQL access, datasets, labeling queues, browser session recording, agent debugger, MCP access, and Slack and email alerts. The homepage also highlights Slack alerts for Signals.
Quick Facts
- Category
- Developer Tool
- Primary use
- AI agent observability and evaluation
- Hosting model
- Open source, self-host anywhere, and on-premise enterprise
- Pricing
- Free plan plus Hobby, Pro, and Enterprise tiers
- Source domain
- lmnr.ai
- Notable workflows
- Tracing, Signals, issue clustering, evals, debugger sessions