Question 1

Does Memory Notch work offline?

Accepted Answer

Recording and transcription do. They run on your Mac's Neural Engine, and once the models are on disk a recording transcribes with the network off. The app itself is not entirely offline: the models download once, and a signed release checks a static URL for a version number daily, which Settings can switch off.

Question 2

Does my audio leave my Mac?

Accepted Answer

No. Audio, transcripts and notes are written to ~/memory on your machine, and none of them is ever uploaded. There is no account and no telemetry. The two things that do use the network carry nothing about you: the models, downloaded once, and a daily version-number check you can switch off.

Question 3

What is speaker diarization?

Accepted Answer

Diarization separates a recording into speaker turns so the transcript shows who said each part. Memory Notch labels turns [speaker 1] and [speaker 2] on your Mac. Two people sharing one laptop microphone is the hardest case and sometimes comes back as a single speaker.

Question 4

Can Claude or ChatGPT read my notes?

Accepted Answer

Notes are plain Markdown files in ~/memory. Any coding agent or local AI tool that can read a directory can read them. There is no MCP server to run, no export step and no API: you give the tool the path.

Question 5

What does Memory Notch need to run, and how much space?

Accepted Answer

macOS 26 or later on an Apple Silicon Mac. A notch is optional: without one the panel is a capsule under the menu bar. The models take about 1.7 GB on disk once every feature has been used, and recordings are stored separately in ~/memory.

Question 6

Can it capture both sides of a call?

Accepted Answer

Yes. Recordings capture system audio alongside the microphone, so the other person is in the transcript too. That needs macOS Screen Recording permission, and the display has to stay awake for system audio to keep arriving.