Lmnr
Laminar - open-source observability platform purpose-built for AI agents. YC S24.
Last verified:
What is Lmnr?
Lmnr is an open-source observability platform purpose-built for AI agents, enabling developers to trace every LLM call, tool execution, and custom function. It goes beyond basic tracing by helping teams extract insights from captured data during development and in production. The platform allows developers to rollout and replay agents from different parts of captured traces, significantly speeding up the development process, while in production it lets users describe patterns and errors in natural language to track them across all traces.
Key features include comprehensive tracing with first-class support for AI SDK, LangChain, Browser Use, and Stagehand; Signals that detect outcomes and failures in plain language and produce structured events; robust Evals to run evaluations against datasets locally or in CI; Datasets & Labeling to build evaluation datasets from production traces; a Playground to replay any traced span and swap prompts or models; and a Platform with real-time traces, full-text search, SQL access, and custom dashboards. The Agent Debugger is a first-of-its-kind feature that lets developers rerun long-running agents from a checkpoint without leaving the page.
Laminar is designed for AI agent developers, machine learning engineers, and teams building production AI applications. It serves companies like Browser Use, OpenHands, Rye, and many others who need to see what their agents are doing, understand why they fail, and spot patterns across millions of runs. The platform is fully open-source under Apache 2.0 license, written in Rust for fast performance, and supports self-hosting via Docker or Helm charts.
Lmnr pricing
Pricing model: Freemium
Free: $0/month with 1 GB data (no overage), 1,000 Signals steps processing (no overage), 15-day retention, 1 project, 1 seat, and community support. Hobby: $30/month with 3 GB data included (then $2/GB), 5,000 Signals steps included (then $0.0075/step), 30-day retention, unlimited projects, unlimited seats, and email support. Pro: $150/month with 10 GB data included (then $1.50/GB), 50,000 Signals steps included (then $0.005/step), 90-day retention, unlimited projects, unlimited seats, and Slack support. Enterprise: Custom pricing with custom limits, on-premise deployment, unlimited projects, unlimited seats, and dedicated support.
Lmnr pros
- Open-source with Apache 2.0 license for full control
- One-line tracing setup for AI frameworks and SDKs
- First-class support for AI SDK, LangChain, Browser Use, Stagehand
- Agent Debugger lets you rerun from any step with full context preserved
- Signals detect patterns in natural language across millions of traces
- Browser session replay synced with agent traces for browser agents
- Full SQL access to query all platform data via UI and API
- Custom dashboards builder with SQL query support
- Run evals locally or in CI to catch regressions before shipping
- Self-host anywhere with Docker in three lines or Helm charts
- Written in Rust for blazing fast performance (25 MB/s)
- HIPAA compliant and SOC2 Type 2 in observation
- Full-text search across span inputs, outputs, and attributes
- Live tracing to watch traces update as spans arrive
- AI-powered trace summarization and error explanation
- Unlimited projects and seats on Hobby plan and above
- No overage on Free tier for data and Signals steps
Lmnr cons
- Free tier limited to 1 GB data and 15-day retention
- Hobby plan at $30/month may be expensive for individuals
- Signals steps incur overage charges after included amount
- Data overage costs $2/GB on Hobby plan
- SOC2 Type 2 compliance still in observation, not complete
- Self-hosting requires technical expertise to set up and maintain
- Email support only on Hobby plan, no phone support
- Enterprise plan requires custom pricing discussion
Frequently asked questions about Lmnr
What is Laminar and what does it do?
Laminar is an open-source observability platform purpose-built for AI agents. It traces every LLM call, tool execution, and custom function, then helps you extract insights from that data. During development, you can rollout and replay agents from any part of the trace. In production, you can describe patterns and errors in natural language and track them across all traces to measure real behavior.
How do I get started with Laminar?
You can get started with Laminar in multiple ways. The fastest is using Skills for reviewable diff in minutes with no manual setup via npx skills add lmnr-ai/laminar-skills. You can also install manually with pip install 'lmnr[all]' or npm install @lmnr-ai/lmnr. After installation, you add Laminar to your app to start capturing LLM and tool spans, then choose between Laminar Cloud or self-hosting.
What AI frameworks does Laminar support?
Laminar has first-class support for AI SDK, LangChain, Browser Use, Stagehand, Claude Agent SDK, LiteLLM, OpenHands SDK, and more. With a single line of code, the Laminar SDK auto-instruments the vast majority of AI frameworks and SDKs, capturing every LLM call, tool use, and function execution.
What is the Agent Debugger feature?
The Agent Debugger is a first-of-its-kind developer experience that lets you rerun long-running agents from a checkpoint without leaving the page. When an agent fails 15 minutes into a run, you can go to the exact step where things went wrong, tweak your prompt or tool definitions, and rerun from that step with full context preserved. The debugger mocks all prior LLM calls by replaying their original responses, so the agent walks through prior steps instantly without spending tokens.
How do Signals work in Laminar?
Signals let you describe outcomes and failures in plain language. Laminar reads every trace and produces structured events you can query, cluster, and alert on. You provide a prompt in plain language describing what the Signal should detect, a JSON schema for the structured output, and optional triggers. Laminar's AI agent detects events from traces based on your definition, then clusters all events for high-level insights.
Can I self-host Laminar?
Yes, Laminar is fully open-source under Apache 2.0 license and can be self-hosted anywhere. You can set it up with Docker in three lines (git clone https://github.com/lmnr-ai/lmnr, cd lmnr, docker compose up -d) or deploy anywhere with Helm charts. Self-hosting gives you full control over your data and infrastructure.
What is the Free tier limit?
The Free tier includes 1 GB of data per month with no overage charges, 1,000 Signals steps processing with no overage, 15-day data retention, 1 project, 1 seat, and community support. This is enough for getting started and lightweight usage without any cost.
How does pricing work for overage?
On the Hobby plan, additional data costs $2 per 1 GB after the included 3 GB, and additional Signals steps cost $0.0075 per step after 5,000 included. On the Pro plan, additional data costs $1.50 per GB after 10 GB included, and additional Signals steps cost $0.005 per step after 50,000 included. The Free tier has no overage charges.
What compliance certifications does Laminar have?
Laminar is HIPAA compliant and SOC2 Type 2 is in observation. This makes it suitable for enterprises that need to meet strict compliance requirements when deploying AI agents in production, especially in regulated industries like healthcare.
How does Laminar handle browser agents?
Laminar captures browser screen recordings and automatically syncs them with agent traces, providing session replay for browser agents. This feature easily integrates with Browser Use, Stagehand, Playwright, Kernel, Browserbase, and more. Developers can see exactly what the agent saw at every step when debugging browser-based AI agents.