[langfuse-bot]

Generated from langfuse/langfuse main at 618aac2.

Confidence Score: 4/5

The import-placement requirement must be satisfied before merging; the duplicate Topics error branches are non-blocking cleanup.

Summary

Updates the generated API clients from the upstream API definition.

• Python callers can inspect Topics facets and topic assignments.
• Organization admins can manage LLM connections shared across projects.
• Score-config names can be up to 80 characters.
• Project deletion docs explain when the AI Gateway turns off.

_{Reviews (1) · Last reviewed commit: "feat(api): update API spec from langfuse..." · Reviewed by Greptile}

[github-actions]

@claude review

[claude]

Claude Code Review

No review was started: this request came from a bot account. Manual reviews can only be requested by someone with write access to this repository. Ask a maintainer to comment @claude review, or have your automation post the comment from a user account with write access.

_{Tip: disable this comment in your organization's Code Review settings.}

[claude]

LGTM. This is a mechanical Fern regeneration of the API client, and it follows the patterns already used in langfuse/api.

What was reviewed:

• The new unstable topics sub-client. Sync and async clients both call the same GET paths (api/public/unstable/topic-facets[/{id}] and api/public/unstable/topics[/{id}]). The client is wired in as a lazy property on UnstableClient and AsyncUnstableClient. The __init__ dynamic-import map and __all__ entries match.
• The organization-level llm_connections methods (list, upsert, delete) under api/public/organizations/llm-connections. Sync and async versions match, and they mirror the existing project-level methods.
• Docstring-only edits: the score config name limit (35 to 80 characters, with ScoreConfig.name now a pydantic.Field) and the note on AI Gateway project deletion.

Extended reasoning...

The change touches 26 Fern-generated files under langfuse/api, adding about 3,800 lines. It adds the unstable topics client and types plus organization-level LLM connection endpoints, and updates some docstrings. There is no hand-written logic and no auth, crypto or permission code. The new methods are thin HTTP wrappers around server endpoints, built the same way as the existing generated clients. The bug hunt ran until it stopped finding anything (dry_streak) and reported nothing. CODEOWNERS covers only .github/, and the PR timeline has no reviewer objections.