Y0 Audio
researchSpeech in, structure out — meetings and voice notes as first-class context.
[ 01 ]Spec sheet
Y0 Audio is the research track that turns the spoken layer of work — meetings, calls, voice notes dictated while walking — into context the runtime can actually use. Transcription is the table stakes; the research problem is everything after it. Who committed to what, by when? Which of the four people talking is the client, and which decision reversed the one made last Tuesday? Y0 Audio produces diarized, speaker-attributed transcripts and then a second structured pass: decisions, action items, open questions, each linked to the moment in the recording where it happened and written into the context graph alongside your documents and calendar. The trust posture is deliberately strict because audio is intimate data: recordings are processed within your project boundary, never used for training, and the kernel logs every access to a transcript the way it logs every document fetch. The family is in research status — quality on clean single-speaker audio is strong, multi-speaker meetings in noisy rooms are an open frontier, and access is limited to research-program partners who can tolerate model revisions between releases. It graduates to preview when diarization accuracy clears the internal bar published on the benchmarks page.
[ 02 ]Capabilities
Diarized transcription with speaker attribution across 4-hour sessions
Decision and action-item extraction linked to timestamps
Voice-note capture that files structure into the context graph
Cross-meeting recall — query what was said across recordings
[ 03 ]Best for
Meeting-heavy teams that lose decisions between calls
Founders and operators who think out loud into voice notes
Research partners building communication-lane prototypes
[ 04 ]Sample request
{
"model": "y0-audio",
"prompt": "Transcribe, diarize, and extract action items.",
"files": ["file_standup_0611.m4a"],
"output_schema": { "type": "meeting_minutes" },
"max_steps": 4
}