Memory Notch
CLOUD SCRIBES
Memory Notch vs Granola
A Mac notepad that captures the call without a bot, then transcribes it in the cloud.
| Memory Notch | Granola | |
|---|---|---|
| Where the audio goes | Nowhere | To a cloud transcription provider, then deleted |
| Processing | On the Neural Engine | Cloud, with local capture |
| Bot in the call | No | No |
| Speaker labels | Yes, on-device | Yes, in the cloud |
| Platform | macOS 26 | macOS, Windows |
| Open source | No | No |
| Price | Not released yet | Free tier, Pro $18/user/month, Business $29/user/month |
Details checked September 2026 against Granola's own documentation. Prices move; treat the figure as a starting point, not a quote.
What Granola does well
- No bot in the participant list. It hears the call the way you do.
- The notes you type are merged with the transcript, so the output reads like yours.
- Audio is deleted after transcription rather than kept indefinitely.
What it costs you
- The audio still leaves the machine: it is passed to a cloud transcriber live.
- Transcripts and notes are retained on AWS in the United States.
- No EU or UK data residency at the time of checking.
Where Memory Notch is different
Memory Notch does the transcription and the speaker separation on your Mac's Neural Engine. There is no account, no telemetry and no API key, and a recording is never sent anywhere. The models download once, and after that a recording transcribes with the network off. Each one becomes a directory of plain Markdown in ~/memory, which means the AI tool you already use can read it by path: no MCP server, no export step, no API.
It is also a smaller thing than Granola. One Mac, one person, no team workspace and no integrations. If you need a shared archive or a CRM, Granola is the better tool and this page is not trying to talk you out of it.