// the find
pydantic/logfire
AI observability platform for production LLM and agent systems.
Logfire is the Python SDK for Pydantic's commercial observability platform, built on top of OpenTelemetry. It wraps OTel with a nicer API and adds first-class integrations for FastAPI, Pydantic models, SQLAlchemy, and a growing list of LLM providers. Target audience is Python teams shipping LLM-powered apps who want structured tracing without the OTel boilerplate.
The OTel foundation means you can eject to Jaeger, Grafana, or any other OTLP-compatible backend — you're not fully locked in to the Pydantic platform. Auto-instrumentation for the common Python stack (FastAPI, asyncpg, httpx, SQLAlchemy) is genuinely good and works out of the box. LLM-specific integrations cover Anthropic, OpenAI, LangChain, LlamaIndex, and PydanticAI with token usage and prompt capture in traces, which is the actually useful part for AI workloads. The SQL query interface for trace data is a smart call — teams that already know SQL don't have to learn another query language.
The backend is closed source, so the SDK is only half the picture — you're evaluating a vendor relationship, not just a library. Self-hosting requires an enterprise license with no public pricing. The free tier limits mean serious production usage will cost money, and there's no way to run the full stack locally for development without paying. Documentation leans heavily on screenshots of the proprietary UI, which tells you nothing about what's actually happening in the SDK internals.