Why notes, and not just a transcript
A transcript is a record, not an output. Nobody re-reads 4,000 words to find out whether the launch date was agreed. The useful artifact is much smaller: what was decided, what someone owns, and what is still open.
That's the split this tool makes. Decisions are kept separate from proposals, because 'we should probably' and 'we agreed' are very different things to circulate afterwards. Action items are attributed to the speaker who volunteered, so the follow-up has a name attached rather than a passive 'it was agreed that'.
Nothing leaves your browser
Meeting transcripts are among the most sensitive text a company produces — pricing, personnel, roadmap, legal. Most AI notetakers ask you to upload exactly that to a third-party server, and many also require a bot to sit in the call as a visible participant.
This tool takes neither. The extraction runs locally in JavaScript, and Smart mode runs a small language model on your own GPU via WebGPU. There is no request carrying your transcript anywhere, which is also why it keeps working on an internal call you would never paste into a cloud tool.
Getting a transcript in the first place
Zoom, Teams and Google Meet can all produce a transcript from a recorded call — usually under the recording or captions menu — and that text pastes straight in here.
If you'd rather not record through the meeting platform at all, that's the gap OmniCanvas fills: it captures the audio on your own machine and transcribes it locally, so no bot joins the call and no vendor receives the recording. The notes then land on the same canvas as the rest of your work.