lmnr-ai/lmnr

What it solves

Laminar is an observability platform designed specifically for AI agents. It helps developers track, debug, and evaluate the performance of their agents by providing deep visibility into the internal steps of an agent's execution flow, which is often a "black box" for developers.

How it works

Laminar provides SDKs for TypeScript and Python that allow developers to instrument their code with a single line of initialization and decorators/wrappers (like @observe()). It is OpenTelemetry-native and integrates with popular LLM providers (OpenAI, Anthropic, Gemini) and frameworks (LangChain, Vercel AI SDK, Browser Use).

Beyond tracing, it includes a set of tools for monitoring and improvement:

• Signals: Allows developers to define behaviors to track in plain English (e.g., "agent is stuck in a loop") and receive Slack notifications.
• Evals: An extensible SDK and CLI for running evaluations locally or in CI/CD pipelines, with a UI for comparing results.
• Data Annotation: A UI for creating datasets for evaluations based on real-world traces.
• Querying: Support for SQL queries to investigate traces, spans, and metrics via a CLI or MCP access for coding agents.

Who it’s for

AI engineers and developers building agentic workflows who need to monitor their LLM calls, debug complex agent loops, and and evaluate the quality of their outputs.

Highlights

• High Performance: Written in Rust with 20x trace compression for efficient storage.
• Broad Integration: Automatic tracing for major LLM providers and agent frameworks.
• Real-time Visibility: Custom engine for viewing traces as they happen.
• Self-hostable: Can be deployed via Docker Compose for local or production use.