v0.11.0

v0.11.0

This release replaces the previous OpenAI docs-derived enrichment flow with a curated static OpenAI enrichment source.

Highlights

• Added openai-static, a checked-in static enrichment source layered on top of live OpenAI inventory from /v1/models
• Added reusable OpenAI static manifest profiles, family defaults, and per-model / per-exposure overrides
• Narrowed OpenAI static enrichment to maintained text models only
• Removed placeholder OpenAI default exposures
• Added OpenRouter-backed audit checks to flag likely drift in OpenAI static enrichment
• Regenerated catalog.json with the new OpenAI static source

Why this changed

The previous OpenAI enrichment path relied on docs-derived fixture data, which was brittle and harder to maintain confidently over time.

OpenAI’s maintained model set is small enough that a curated static manifest gives us a better balance of:

• accuracy
• explicit provenance
• low maintenance overhead
• precise per-model overrides
• less guesswork

What this means for consumers

OpenAI data is now split more clearly into:

• openai-api: live inventory from /v1/models
• openai-static: curated enrichment for maintained text models

As a result:

• maintained OpenAI text models now carry explicit openai-responses exposures with normalized parameters, mappings, and parameter values where curated
• non-text OpenAI inventory (audio, realtime, image, transcription/TTS, moderation, etc.) may still appear in the catalog, but no longer receives broad static exposure enrichment by default

Internal maintenance improvements

• Added maintainer guidance under internal/source/openai/README.md
• Added OpenRouter-backed audit coverage to help spot likely drift and missing high-confidence capability metadata
• Renamed the build override flag from --openai-docs-dir to --openai-static-file

Notes

This release intentionally keeps OpenAI static enrichment conservative: when a capability is uncertain, it is omitted rather than guessed.