Meeting transcription with speaker context and evidence you can open

Oat does not need to join a call as a participant. After you press Record Meeting, ScreenCaptureKit captures the audio produced by the selected meeting application or display, while the microphone records your side as a separate stream. Video is neither captured nor retained.

1. 01

   Start deliberately

   Choose Record Meeting, give the recording a title, and select the computer-audio source. Oat never begins because a calendar event started or a call app opened.

2. 02

   Keep channels separate

   Your microphone and the selected computer audio are captured as separate timestamped streams. The microphone establishes which speech belongs to the local user.

3. 03

   Transcribe speech chunks

   Voice activity detection closes completed speech chunks and sends them through the local recognizer while the meeting continues, building an ordered transcript over time.

4. 04

   Review and name

   After the recording, correct uncertain speaker labels and review important names, numbers, decisions, and tasks against the cited transcript and retained audio.

System audio capture requires Screen & System Audio Recording permission; microphone capture requires Microphone permission. Calendar information may suggest a meeting title or participant names when a future connector is configured, but it does not start a recording. The current download does not enable Google or Slack connectors.

FluidAudio’s voice-activity detection identifies completed speech chunks rather than waiting for the entire meeting to end. The local English Parakeet speech recognizer transcribes those chunks, and Oat orders them using the stream timestamps. Processing time can vary with meeting length, speech patterns, audio quality, and Mac load.

Recording continues while the transcript is assembled. If you lose the thread, What did I miss? uses the preceding five minutes of finalized transcript to organize the recent topics, decisions, questions, and action items without stopping capture.

Separating microphone and computer-audio channels solves one important attribution problem: Oat knows which speech came from the local microphone. For mixed remote audio, diarization estimates where one speaker ends and another begins. It aligns those regions with recognized words before merging them into the meeting timeline.

Oat keeps confirmed voice profiles locally. When a later speaker embedding is similar enough to a remembered profile, it can suggest that person’s name. If the match is below a conservative threshold, the transcript keeps a neutral label such as Speaker 1 and asks for confirmation rather than attaching a confident but unsupported identity.

You can correct a label and remember that mapping for future meetings. Similar voices, overlapping speech, poor audio, or a different microphone can still produce mistakes, so attribution should be checked before assigning an owner to a decision or task. For the broader notes workflow, see Oat AI meeting notes.

A transcript segment is more than displayed text. It includes a meeting ID, start and end times, audio channel, speaker or voice-profile reference, raw and corrected text, and a stable source ID. Those IDs let MumbleFlow attach evidence to downstream notes.

Selecting a citation opens the meeting at its timestamp and can play the corresponding local audio while the recording is retained. If the source does not support an answer, MumbleFlow’s grounded search is designed to say it could not find the information rather than fabricate an uncited response. Learn how retrieval works on the local AI search page.

Meeting recordings are encrypted on the Mac using a key protected through the macOS Keychain. Audio expires after 30 days by default, and Settings exposes storage use plus immediate deletion controls. The transcript and cited notes can remain after audio deletion, but their citations can no longer play a recording that no longer exists.

Speaker profiles, transcript text, meeting notes, and the local search index also remain on the device until their applicable deletion controls are used. MumbleFlow has no product account, content server, cloud sync, ambient recording, or automatic meeting start in this release.

Read the privacy, encryption, and retention details, compare all meeting and dictation capabilities on the features page, or use the support guide to troubleshoot system-audio routing.