Laminar

Laminar is an open-source observability platform for AI agents, built by LMNR AI. Once you add its SDK, each agent run is recorded as a trace with the model inputs and outputs, tool calls and costs, and the README says one line of code can auto-trace common frameworks and models such as Vercel AI SDK, Browser Use, LangChain, OpenAI, Anthropic and Gemini. Its most distinctive feature is Signals: you describe a behavior you want to watch for in plain English, such as "agent is stuck in a loop", and Laminar uses AI to read every run and pings you in Slack when it happens, which helps catch runs that went wrong without throwing an error. It also has an evals toolkit, dashboards you can query with SQL, data annotation and dataset tools, and MCP and CLI access so a coding agent can query your traces. You can use the hosted service or self-host it with Docker Compose.

Who it's for:Developers and teams building AI agents who want to see what each step did and where things quietly went wrong. Readers who don't write code won't have a use for it.

Before you start: The hosted service needs an account and a project API key, and then you add the SDK to a Python or TypeScript project. With the hosted service, your agent traces (including prompts, model inputs and outputs, and tool results) are uploaded to Laminar, and optional AI features also send related data to Google Cloud or AWS Bedrock. The privacy policy says Privacy Mode is off by default on the Free and Starter plans, and while it's off, run data processed by Signals may be used to train Laminar's own models; on Pro and above it's on by default. AI Inspiration World doesn't recommend sending traces that contain personal data or secrets to the cloud: self-host, or turn on Privacy Mode before you connect. Self-hosting means maintaining the database and services yourself, and if you connect an outside model, data still goes to that model provider. The hosted service has a free plan and paid subscriptions, Signals isn't included in the free plan, and you should check the official site for prices. The code is licensed under Apache-2.0, and the hosted service has its own terms.

Based on the official site and the project's public documentation. Not tested by AI Inspiration World.