Drop in an audio recording and the browser transcribes it for you, or paste a transcript you already have. Either way you get organised minutes back: a summary, the decisions, and action items with owners where the meeting named them. Recordings are capped at 15 minutes to keep the tool free for everyone.
or paste a transcript
Recordings are capped at 15 minutes so the tool stays free for everyone. For a longer meeting, split the file or paste the transcript.
The quality of the notes depends almost entirely on the quality of the transcript you feed in.
Direct answer: useful meeting notes preserve decisions, owners, deadlines, and uncertainty from the transcript; they do not infer attendance, invent an owner, or turn a suggestion into a commitment. Run the fixture below before trusting notes from a real meeting.
Intent fence: this page turns supplied audio or text into minutes. It does not join calls, identify speakers, record consent, or replace review of the source transcript.
Maya: We will keep the October launch date. Dev, please send the revised onboarding copy by Friday, October 23. Dev: I can do that. Lina: The billing migration is still blocked because Finance has not approved the tax mapping. I will ask Finance today, but there is no delivery date yet. Maya: Do not enable automatic refunds in production. We can test them in sandbox after Legal signs off. Dev: Should I also update the cancellation email? Maya: Let us decide that next week. The only approved change today is the onboarding copy.
| Type | Expected output | Reject if |
|---|---|---|
| Decision | Keep the October launch date | A new date is invented |
| Action | Dev — revised onboarding copy — Oct 23 | Owner or deadline is omitted |
| Action | Lina — ask Finance about tax mapping — today | A migration delivery date is invented |
| Guardrail | No production auto-refunds; sandbox only after Legal approval | The conditional test becomes approved production work |
| Open item | Cancellation email decision deferred to next week | It is presented as an assigned task |
Pass rule: 5/5 rows correct, zero invented owners or dates, and every action traceable to an exact sentence. If the transcript has no speaker label, the output should say “owner not stated,” not guess.
Privacy boundary: obtain recording consent under the rules that apply to the participants. Audio is decoded in the browser and sent in short transcription segments; the corrected transcript is then sent for note generation. Do not submit confidential material unless that processing path is acceptable to your organization.
MP3, M4A, WAV, OGG, Opus, FLAC and most video containers. Your browser decodes the file, so anything it can play will work.
Transcription costs money on every call, and the tool is free with no account. A cap keeps one long upload from spending the budget everyone else shares. For a longer meeting, split the recording or paste a transcript instead.
The audio never gets uploaded as a file. Your browser decodes it, converts it to 16 kHz mono, and sends short segments for transcription. Nothing is stored on our side, and closing the tab discards everything.
Good on clear speech, weaker on names, acronyms and crosstalk. Read the transcript before you generate notes, because a misheard name propagates into the action items.
No. It returns a single transcript without speaker labels. If you need attribution, add the names yourself before generating notes, and the action items will pick them up.