Platforms¶
Each pool publishes exactly one entity in the matching Home Assistant domain.
ai_task¶
Attachments are not supported. They arrive already resolved and there is no supported way to pass a resolved attachment to another entity, so Home Assistant rejects such calls before they reach the pool.
conversation¶
A member answering with an error response counts as a failure, otherwise the pool would return "sorry" from the first broken member and never reach a working one. Languages are advertised as match-all.
tts¶
Languages and options are the union across members; a member that cannot handle a request raises and the next one is tried.
The pool entity is named Text-to-speech under the pool's device, so a new
pool's entity id is tts.<pool>_text_to_speech. The other pool types take the
device's name, but the TTS manager refuses an engine whose entity.name is
None. Pools created before this change keep their entity id.
stt¶
Audio is buffered so a second member can be given the same recording. The buffer defaults to 8 MB and is set in the members step of the UI (megabytes in the form, bytes in storage).
Audio format capabilities are the intersection across members, because the pipeline encodes once before any member is chosen. A recording larger than the buffer is clipped to what fits and gets one attempt with no failover: handing the same half-sentence to a second member cannot produce a better transcript. The log warning says so.