logfire

Why we included this project

Production LLM and agent systems tend to fail in the seams: a slow vector search, a timed-out tool call, or a costly prompt. Standard APM tools rarely show those AI-specific layers, and that's the gap Logfire fills. It pulls traces, metrics, and logs from LLM calls, agent reasoning steps, database queries, and API requests into a single timeline, and because it wraps OpenTelemetry, instrumentation you already have for FastAPI, LangChain, OpenAI, and other libraries keeps working. There are SDKs for Python, JavaScript/TypeScript, and Rust, plus extras plain logging tools skip: real-time cost tracking, budget alerts, and an evaluation framework for catching regressions on production traces. If you are debugging why an agent misbehaves or where latency and spend come from, this gives you one place to look instead of stitching together separate monitoring tools.