Pydantic Logfire

Questions from the bench

What is AI observability and why does it matter?

AI observability tells you why something failed by tracing, debugging, and evaluating AI applications across LLM calls, databases, tools, costs, and outputs. It matters because AI applications fail differently from traditional software—models can return plausible but wrong answers, tool calls can misfire silently, and costs can spike from retry loops. Without observability across the full stack, debugging becomes guessing.

Do I need to use Pydantic AI or Python to use Logfire?

No to both. Logfire is framework-agnostic and language-agnostic. It ingests standard OpenTelemetry and works with OpenAI, Anthropic, LangChain, LangGraph, LlamaIndex, CrewAI, and other stacks that emit OTel. TypeScript and Node.js are first-class with SDKs, and any language with an OpenTelemetry SDK works including Go, Java, .NET, Ruby, and Rust.

Does Logfire do evals or just tracing?

Both. Logfire includes a full evaluation suite for building datasets from production traces, running experiments comparing baselines, and scoring with code assertions, LLM-as-a-judge rubrics, or human review. Eval results are stored as queryable telemetry so evaluation and observability share the same data.

Can my coding agent query production telemetry?

Yes. Logfire stores traces as queryable data in standard PostgreSQL SQL, and its MCP server lets a coding agent like Claude Code or Cursor query production directly to debug from what actually happened instead of guessing.

Is Logfire open source?

The Logfire SDK is open source under the MIT license and available on GitHub as a wrapper around the OpenTelemetry Python and TypeScript SDKs. The hosted platform is the commercial product.