How the agent's video reaches the host: a player the driver creates through its creation seam ('playout', the
kit records it), a live MediaStream ('stream', as a WebRTC provider hands over), or never ('none', audio
only).
OptionalEnvironmentChecks this test environment can't run faithfully, by id, with the reason. They are reported skipped; a skip here says nothing about the provider.
Whether this provider's server can grant an avatar at all. The Gemini Developer API can't.
OptionalTimedWhether the model's voice and video share one media timeline that the driver keeps: it queues each PCM chunk with
its media time and hands its voice playback to the video player as the clock (VideoPlayoutOptions.Clock), so the
face moves with the voice. Default false (Gemini's PCM carries no media time).
Whether the avatar's video can carry the voice, so the kit checks the voice never plays twice.
OptionalVideoThe form the model's video comes in, as Core's RealtimeVideoFrame kinds: fragmented MP4 (a Gemini avatar), encoded
chunks, or images. The kit sends frames of this kind. Default 'fmp4'.
Whether the driver reports the seconds of video the model generated (OutputTokenDetails.VideoSeconds).
Who plays the agent's voice: the driver's PCM playback (the kit's recorder), or the transport (WebRTC).
What a provider can do. A check that needs something the provider can't do is skipped with the reason.