The problem
Every AI note-taker on the market works the same way: a bot joins your call or a background service grabs your audio, ships it to a server, and hands back a summary a few minutes later. Your conversation ends up on someone else's infrastructure under someone else's retention policy, and the summary is often mediocre anyway, because running a good model on every user's call is expensive.
thE-noteTaker moves the whole pipeline onto the machine that's already sitting on your desk.
How it works
It lives in the menu bar as a lowercase e. Click it, recording starts, capturing microphone and system audio at the same time so you get both sides of a call. Click it again, recording stops, and transcription kicks off in the background, so the next recording can start right away without waiting on it. Once it's done, the recording shows up in a library with status, retry/re-transcribe options, and inline edits.
Everything lands on disk as plain markdown transcripts and WAV files. No vendor lock-in, no database, and your files keep working even if the app doesn't.
Under the hood
Transcription runs through parakeet.cpp or whisper.cpp, both bundled as native binaries, with optional speaker diarization. There's an optional AI pass for summaries and action items through Ollama, Claude CLI, or Codex CLI. Ollama runs on your Mac. The Claude and Codex CLIs send the transcript text to Anthropic or OpenAI, the same as if you'd typed it in yourself. Audio never leaves the machine, and the whole thing is opt-in: the core record/transcribe loop doesn't need it at all.
Privacy by design
Sensitive recordings can be flagged and are physically relocated on disk, so "excluded from AI context" is enforced by where the file lives, not a checkbox the app has to remember to respect.
Safety
A recording set to Full Isolation is never sent to an AI provider.
By the numbers
By the numbers
199
issues closed of 239
6
milestones
Issues by area
- ui55
- build35
- storage35
- ai34
- meeting-detection20
- security19
- transcription17
- audio-capture17
- integrations12
- speaker-id11
Counted from the repo and its board on Oct 1, 2026, since Jun 26, 2026.
Built with
Swift and SwiftUI. Microphone audio comes through AVAudioEngine and system audio through ScreenCaptureKit, which is why macOS asks for Screen Recording permission (audio only, never your screen). Transcription runs on parakeet.cpp or whisper.cpp. It builds for macOS 14 or later on Apple Silicon and is used day to day on macOS 26 and 27.
It's a solo side project, still rough in places, but the core loop works and I use it daily. The technical writeup and workflow writeup above go into how it's built and how a one-person project stayed organized enough to actually ship.