Live minutes

Minutes are produced section by section while the session is still running.

With digest: true, minutes update after enough new transcript text accumulates (2,000 characters by default; configurable by the service). Updates add new points while retaining earlier sections. At session end, only unprocessed text is summarized, even if below the threshold. Each digest.update carries the complete accumulated sections snapshot.

Request

json

{  "digest": true}

Fields

Field	Type	Description
`digest`	`boolean?`	Stream minutes as the session runs, pushed as `digest.update`. Default `false`.

What you receive

json

{  "type": "digest.update",  "sections": [    {      "status": "confirmed",      "title": "Quantization step and noise floor",      "points": ["Quantization step near zero is about 8", "Roughly 22 dB of headroom"],      "startMs": 0,      "endMs": 252000,      "segmentIds": ["gw_8f3a_1", "gw_8f3a_2"]    },    {      "status": "pending",      "title": "Gain adjustment",      "points": ["Re-measure after lowering the gain by 3 dB"],      "startMs": 252000,      "endMs": 268400,      "segmentIds": ["gw_8f3a_3"]    }  ]}

Sections follow topics, not a fixed template. The current topic stays pending and keeps gaining points; when the topic changes, that section becomes confirmed and a new one opens. startMs / endMs are computed from the transcript segments, not written by the model, so every section maps back onto a position in the recording.