AI Meeting Notes - record, transcribe & summarize meetings locally

What it does

Click Record at the start of a call. When it ends, you get a Markdown note in your vault with a summary, action items, and a speaker-labeled transcript - all generated locally.

🔒 Fully local & private

Transcription runs on your machine via whisper.cpp. Summaries can use a local model (Ollama). Nothing is uploaded unless you opt into the cloud summary.

🎙️ Captures the whole meeting

Records your mic and the other participants’ audio as separate tracks - no virtual audio device or Multi-Output Device to configure.

🗣️ Speaker-labeled transcript

Because the two sides are separate tracks, every line is labeled You or Them - no diarization model needed.

🌍 Any language

Auto-detects and transcribes the spoken language with a multilingual model.

📝 Summary & action items

Generated by a local Ollama model or the Claude API - switchable, with a per-model prompt you control.

🗂️ A meetings library

Browse past meetings, rename them, re-generate the transcript/summary, or delete a meeting and its audio.

How it works

Mic ─┐ ├─► Core Audio capture ─► whisper.cpp ─► diarized transcript Sys ─┘ (no virtual device) │ local LLM (Ollama) or Claude │ Markdown note in your Obsidian vault

System audio is captured with macOS Core Audio process taps (macOS 14.4+), so there’s nothing to route manually. Transcription is batch: record the whole meeting, then process on stop.

Requirements

macOS 14.4 or later (Apple Silicon recommended).
A whisper model - download one from inside the app (Settings → Transcription).
No extra installs - whisper.cpp is bundled inside the app.
Optional, for summaries: Ollama (local) or a Claude API key.

Install & setup

Most people just download and install - no building required.

🍎 macOS

Download the .dmg, open it, and drag AI Meeting Notes to Applications. It's signed & notarized by Apple, so it opens with no Gatekeeper warning.

🪟 Windows beta

Download the installer (or portable zip) and run it. Self-contained - no .NET install needed. Windows 10/11.

Do I need to build it myself? No. The download above is the finished, ready-to-run app - building from source is only for developers who want to modify or audit the code. A signed release is exactly so you can skip building.

First run

Open Settings (macOS: ⌘,).
Choose a folder for your notes. Notes are plain Markdown, so any folder works. If you use Obsidian, point it at your vault (notes go to <folder>/Meetings, audio to .../recordings) and the embedded audio renders as inline players - but Obsidian is optional.
Pick or download a whisper model (start with base); set Language to auto.
Choose a summary engine - local Ollama or Claude API - or None.

The first time you click Record, your OS asks permission to record audio and use the microphone - click Allow for both.

For developers - build from source

Only needed if you're contributing or want to build it yourself.

# macOS app
cd ai-meeting-notes/packages/meeting-engine
./scripts/build-app.sh && open ".build/AI Meeting Notes.app"

# Windows app
cd ai-meeting-notes/packages/meeting-notes-windows
dotnet run --project src/MeetingNotes.App

Using it

Record / Stop: click Record at the start of a call; watch the live System & Mic level meters; click Stop & Transcribe at the end.
Review: the new meeting appears in the sidebar with its summary, action items, and transcript.
Rename: edit the title on the right; the audio stays linked.
Re-generate: re-transcribe & re-summarize with your current settings (model, language, prompt) - great after switching to a multilingual model.
Delete: remove a meeting and its recordings from the detail view or the list’s right-click menu.

Privacy

Recording, transcription, and (with Ollama) summarization all happen on your device. The only time data leaves your Mac is if you choose the Claude API summary engine, which sends the transcript text to Anthropic. Local-only setups make no network requests except downloading a model when you ask.

Prefer Obsidian?

This project also ships as an Obsidian plugin that records and transcribes inside your vault (using a loopback audio device for system audio). See the plugin README and the setup guide.