Laminar

Laminar is an open-source observability platform for AI agents. It provides developers with the means to collect and analyze data to improve the quality of LLM applications through tracing and monitoring. The platform focuses on providing visibility into the internal workings of AI agents to help developers identify bottlenecks and failures in their logic.

The software is deployed as a self-hosted server using Docker Compose or via a managed platform. It integrates with applications through TypeScript and Python SDKs, allowing developers to instrument their code with a few lines of configuration. This setup enables the automatic capture of inputs and outputs from various LLM providers and frameworks, sending the telemetry data to the Laminar backend for visualization and analysis.

Key features

• OpenTelemetry-native tracing for Vercel AI SDK, OpenAI, Anthropic, and Gemini
• Signal tracking to monitor specific agent behaviors with Slack notifications
• Extensible SDK and CLI for running evaluations locally or in CI/CD pipelines
• SQL-based querying of traces, spans, metrics, and events via MCP or CLI
• Custom dashboard builder for visualizing metrics and events using SQL
• Data annotation UI for creating evaluation datasets
• Trace compression and a realtime engine written in Rust
• Full-text search over span data and gRPC exporter for tracing data

The backend is written in Rust to ensure high performance and efficient storage through 20x trace compression. It uses a PostgreSQL database for data persistence and supports custom schemas for shared database environments. The platform includes a custom realtime engine for viewing traces as they happen and a dedicated UI for comparing evaluation results. It is designed for developers building complex AI agents who require deep visibility into execution flows and the ability to run systematic evaluations across different model versions.

Laminar is a specialized LLMOps tool focused on the observability and evaluation lifecycle of AI agents.