From Home, select a file and choose:
- Quick transcript for text only.
- Meeting mode for text plus optional offline speaker separation.
The upload is decoded before a durable job is accepted. Task progress appears under Tasks. A completed transcript can still report speaker separation as unavailable; this is not treated as an identity result.
Open Meetings, search for transcript text, and select a result to jump to its exact segment. Double-click text to edit it. Anonymous speaker labels can be reassigned in batches. Confirm a person only after listening to the audio.
Create a person under People and upload a clear voice sample of at least two seconds. Expand Manage samples to preview recordings, inspect effective-speech and quality indicators, or remove a mistaken sample. Samples and embeddings remain on the local machine. A model-engine change can make existing embeddings incompatible; add a new sample for the active engine if matching is unavailable.
Live mode stores audio and final transcript segments as a meeting. Captions and speaker labels are approximate. Use an uploaded full recording when post-meeting quality matters. Browsers require microphone permission; non-loopback mobile access generally requires HTTPS.
Summary, actions, and minutes have a local extractive fallback. Enabling an external LLM sends transcript text to the configured endpoint. Review generated notes before use. Export formats are Markdown, SRT, WebVTT, and JSON.