A pre-release for field validation. HACS offers this only to users who have enabled beta versions for this repository; it is not a public release and the changes below still ship in the next monthly bundle as 3.43.0.
Cut so @yagui75 can validate the fix for #648 against Chatterbox-TTS-Server, which no maintainer can reach.
What to validate
Adding a TTS Provider → Local (OpenAI-compatible) pointed at a TTS-only server should now complete instead of failing with "Can't reach the provider", and the resulting TTS entity should speak through an Assist pipeline.
The fix (#650, #648)
The local STT/TTS credentials step proved a server was reachable by asking it for its model catalog (GET {base}/v1/models). Every general-purpose server implements that; a single-purpose TTS server has no reason to, and Chatterbox-TTS-Server answers 404 there while serving /v1/audio/speech correctly — so setup was impossible against it, with nothing logged even at debug level.
When the catalog is missing, HGA now probes the route the provider will actually use — /v1/audio/speech for TTS, /v1/audio/transcriptions for STT — and accepts the server when that route is served. A URL that serves neither is still rejected, so a typo is still caught; a server that rejects your API key is still reported as bad credentials rather than as unreachable; and providers that pass no capability path (chat and embedding model providers, the startup health check) are unchanged, so nothing about their validation is loosened.
Everything else already on main since v3.42.0 is included. See CHANGELOG.md under Unreleased.