Pydantic Logfire

license

MIT(covers the Python/TypeScript/Rust SDKs and documentation in this repository only)

source

partial(the SDK is published

backend

commercial(hosted Logfire Cloud platform, with self-host offered only under a purchased enterprise license)

The repository is a genuine, standalone MIT SDK that exports OpenTelemetry data to any compatible backend, but the running platform - the UI and the server that records and displays it - is closed source by the vendor's own statement, with self-hosting sold as a paid enterprise product rather than published. That is the `partial` rung regardless of the SDK's license, the same reading `weave` gets for the same shape, though Logfire publishes less of the server side than weave does - no trace-server library ships here at all.

• https://raw.githubusercontent.com/pydantic/logfire/main/README.md recorded 2026-09-02

  Two statements read directly rather than assumed. First: "This repo contains the Python SDK for `logfire` and documentation; the server application for recording and displaying data is closed source." Second, under "Logfire Open-Source and Closed-Source Boundaries": "The Logfire SDKs ... are open source ... The Logfire platform (the UI and backend) is closed source. You can self-host it by purchasing an enterprise license."

• https://raw.githubusercontent.com/pydantic/logfire/main/LICENSE recorded 2026-09-02

  MIT License text, Copyright Pydantic Services inc., covering the contents of this repository (the SDK and docs).

The scored product is the closed hosted platform, and nothing counts its use. The vendor page names a roster of paying customers - Motorola, Sophos, Weaviate, Nous Research, Airbyte, Vox Media, Pictet and about fifteen more - and publishes no user, account or trace-volume figure, so the level rests on that standing as a real, mid-sized commercial observability platform rather than on a measurement. The logfire package's 13,695,048 PyPI downloads a month are deliberately not the band. The SDK exports to any OpenTelemetry backend and is pulled in as a dependency by Pydantic AI and other instrumentation, so an install of it is not a use of the platform, and reading a level 5 off it would credit the platform with the whole Pydantic ecosystem's install base. The repository's 4,450 stars are consistent with a level of this size and would band lower on the stars scale.

• https://pydantic.dev/logfire recorded 2026-09-02

  "Companies who trust Pydantic Logfire" lists Boosted.ai, DeepScribe, Airbyte, Evergreen.ai, Fleet AI, Seekr, Polar.sh, Stuut, Alpaca, Motorway, WorkWhile, Sophos, Aignostics, Amboss, Motorola, Epistemix, Nous Research, Pictet, SimpleClub, Tiger Data, Vox Media, ZenHub and Weaviate, plus one Dosu case study. The only figure on the page is the free-tier quota ("10M spans, logs, and metrics free every month"); no user count, customer count or usage volume appears anywhere.

• https://pypistats.org/api/packages/logfire/recent recorded 2026-09-02

  last_month = 13,695,048 downloads for logfire (last_week 2,817,073, last_day 559,383). A measurement of the SDK channel, recorded as context and explicitly not the basis of the band.

• https://api.github.com/repos/pydantic/logfire recorded 2026-09-02

  stargazers_count = 4,450, forks_count = 283, pushed_at 2026-09-02, for pydantic/logfire.

Covers every dimension this category grades: tracing, prompt management with versioning and rollout, a playground, and evaluation through datasets, experiments, live evals and human review - plus SQL access over the whole recorded corpus and deep Pydantic-specific instrumentation that the anchor does not have. That puts it level with Langfuse at 4.

• https://raw.githubusercontent.com/pydantic/logfire/main/docs/navigation.yml recorded 2026-09-02

  Sidebar listing a whole "Prompt Management" section (concepts, UI, templates, composition walkthrough, promotion and rollouts, scenario testing, application usage) and a separate "Prompt Playground" page, alongside an "Evaluate" section with datasets-and-experiments, live-evals, human-review and annotate-agent-runs pages.

• https://raw.githubusercontent.com/pydantic/logfire/main/docs/ai-observability.md recorded 2026-09-02

  Deep agent-tracing description - one trace covering model calls, tool calls, database queries and API requests in nested spans, with a worked example.