Back to Gap Map Product / UX / Telemetry & observability

Pydantic Logfire

source available / Overall score: 3.5

AI and application observability platform from the Pydantic team, built on OpenTelemetry, providing agent and LLM tracing, prompt management with versioning and rollout, live evaluations, datasets and experiments, and SQL querying over recorded traces, metrics and logs. The Python, TypeScript and Rust SDKs are open source and can export to any OpenTelemetry-compatible backend; the Logfire platform itself - the UI and the server that records and displays the data - is closed source, with self-hosting available only under a purchased enterprise license.

The repository's own README states this directly - "This repo contains the Python SDK for logfire and documentation; the server application for recording and displaying data is closed source" - and its "Open-Source and Closed-Source Boundaries" section confirms the platform (UI and backend) is closed, self-hostable only under a paid enterprise license. The scored product is that platform. The logfire PyPI package is deliberately not declared as an artifact, because installing the SDK does not measure use of the platform - it exports to any OpenTelemetry backend and is pulled in as an instrumentation dependency by Pydantic AI and other libraries, so its install count can move independently of how many people send data to Logfire. The repository stays declared because it is the vendor's own canonical repository for the product (the SDK and the documentation), the openness read is drawn from it, and it is not a third-party binding.

Openness

2 high confidence

2.0

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)

Logfire's SDK is a genuine, standalone MIT package that exports OpenTelemetry data to any compatible backend, but the platform that records and displays it, the UI and server, is closed source, and self-hosting it requires a purchased enterprise license.

https://raw.githubusercontent.com/pydantic/logfire/main/LICENSE recorded 2026-10-07
MIT License text, Copyright Pydantic Services inc., covering the contents of this repository (the SDK and docs).
https://raw.githubusercontent.com/pydantic/logfire/main/README.md recorded 2026-10-07
"This repo contains the Python SDK for `logfire` and documentation; the server application for recording and displaying data is closed source." Under "Logfire Open-Source and Closed-Source Boundaries" the SDKs "are open source", while "The Logfire platform (the UI and backend) is closed source. You can self-host it by purchasing an" enterprise license.

Adoption

3 low confidence

3.0

Logfire's vendor page names a customer roster including Motorola, Sophos, Weaviate and Airbyte but publishes no user, account or trace-volume figure, so the reading reflects that standing rather than a measured count. The SDK's PyPI downloads and GitHub stars measure Pydantic's broader instrumentation ecosystem rather than use of the closed platform, so neither can raise the reading.

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.
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 reading.

Capability

4 high confidence

4.0

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 Pydantic-specific instrumentation that Langfuse does not have. That breadth keeps it level with Langfuse rather than ahead of it.

https://raw.githubusercontent.com/pydantic/logfire/main/docs/ai-observability.md recorded 2026-10-07
Deep agent-tracing description - one trace covering model calls, tool calls, database queries and API requests in nested spans, with a worked example.
https://raw.githubusercontent.com/pydantic/logfire/main/docs/navigation.yml recorded 2026-10-07
Sidebar listing a whole "Prompt Management" section (templates, composition walkthrough, promotion and rollouts, scenarios) and a "Prompt Playground" page, beside an "Evaluations" section with "evaluate/datasets-and-experiments", "evaluate/live-evals", "evaluate/human-review" and "evaluate/annotate-agent-runs" pages.

Verified 2026-10-07, except some scores, still awaiting re-confirmation