Three steps, zero cloud
Tape sits in the menu bar until a meeting starts. From there it records, transcribes and summarizes on your own machine, and leaves the results as plain files you can read with anything.
1. Record
Start a recording however suits you: the global shortcut ⌥⌘R, a click on the menu-bar icon, or by letting Tape notice for you.
Meeting detection watches for one of your meeting apps taking the microphone. When that happens, Tape offers to record and pre-names the session from the calendar event in progress, or from the call window's title. You can set it to ask every time, to start recording automatically, or to stay out of the way entirely.
Two tracks are captured separately: your microphone, and everything the Mac itself plays. The second one uses macOS's own system-audio tap, so there are no kernel extensions and no virtual audio drivers to install. The menu-bar icon carries a red dot while recording.
What you grant, once
- Microphone, to record your side
- System audio recording, to record everyone else
- Calendar, optional, only so a recording can be named after the meeting it belongs to
2. Transcribe and summarize
Stop the recording and the rest happens on its own. Each track goes through an on-device speech model, the two are merged back together by timestamp, and you get a speaker-separated transcript: your microphone is you, the system track is everyone else.
Then a local language model reads the transcript and writes the summary: TL;DR, key points, decisions, action items. The menu-bar dot glows yellow while any of this is running, and the app is perfectly usable in the meantime.
Both steps pick their model based on the language you selected. Parakeet is the fast path for English; Whisper covers the other languages and can detect which one was spoken. More on that on the models page.
Transcription resumes by itself. The queue lives on disk rather than in memory, so quitting the app or losing power mid-transcription costs you nothing: the pending recording is picked up again the next time Tape starts.
3. Use it
Read the transcript and click any line to hear that exact moment. Pin comments to specific seconds. Write meeting notes in Markdown. Search the whole library from the sidebar. Ask the AI what was decided, what is on you, or for a draft of the follow-up email.
Everything you see is already a file in a plain folder: audio, Markdown, JSON. Copy it, sync it, grep it, feed it to your own tools. See where your files live.
What this costs you in machine terms
| Step | Runs on | Rough cost |
|---|---|---|
| Recording | CPU, negligible | Two audio files, about 30 MB per hour |
| Transcription | Neural Engine and GPU | Minutes, not hours, for a typical meeting |
| Summary and chat | GPU, via MLX | The model's own size in memory while it runs |
Models are released and unloaded as soon as the queue is empty, so Tape idles at almost nothing between meetings.
Try it on your next call
Free forever. Your recordings, your Mac, your files.
macOS 15+ · Apple Silicon