Privacy-first AI meeting notes for macOS. Records meetings, transcribes them locally, labels speakers on device, and generates AI notes — all on your machine.
Website: ultra.ashref.tn
- Records meetings from your microphone and system audio simultaneously with automatic ducking.
- Transcribes in real time with local Whisper, Parakeet, or Qwen3 ASR models — raw audio never leaves your machine.
- Labels speakers automatically after each recording with on-device pyannote + WeSpeaker ONNX diarization.
- Summarizes with AI using your choice of provider: local Ollama, OpenAI, Claude, Groq, OpenRouter, or any OpenAI-compatible endpoint.
- Searches across all meetings with full-text transcript indexing.
- Imports existing audio in MP3, WAV, FLAC, OGG, MP4, MKV, WebM, or WMA format.
- Provides menubar quick access with one-click recording start/stop. The tray icon turns red while recording.
Ultra is built on the principle that meeting data is sensitive and should stay on your device.
- Audio and transcripts never touch a third-party server (unless you explicitly choose a cloud AI provider for summaries).
- Transcription and diarization run on device with Metal GPU acceleration on Apple Silicon.
- No telemetry. No analytics. No login. No account required.
- Recordings live in
~/Movies/ultra-meet-recordings/by default. The database lives under~/Library/Application Support/tn.ashref.ultrameet/.
See the full Privacy Policy.
Download the latest .dmg from the releases page.
- Open the
.dmg - Drag
Ultra Meet.appto your Applications folder - Launch from Applications — macOS will prompt once for microphone and screen recording access.
Prerequisites: Rust (stable), Node.js 18+, pnpm, Xcode command-line tools, ffmpeg via Homebrew.
git clone https://github.com/Ashref-dev/ultra-meet-notes
cd ultra-meet-notes/frontend
pnpm install
pnpm run tauri buildOutput: target/release/bundle/macos/Ultra Meet.app and target/release/bundle/dmg/Ultra Meet_<version>_aarch64.dmg.
Ultra does not hard-code a single provider. Configure your preferred model in Settings.
| Provider | Where models run | API key required |
|---|---|---|
| Built-in (llama-helper) | On-device, bundled | No |
| Ollama | On-device, your setup | No |
| OpenAI | Cloud | Yes |
| Claude | Cloud | Yes |
| Groq | Cloud | Yes |
| OpenRouter | Cloud, any model | Yes |
| Custom endpoint | Any OpenAI-compatible | Optional |
The recommended default is Ollama with a small local model such as qwen2.5:3b or llama3.2:3b for fully local summaries at zero API cost.
Ultra is a single Tauri 2 application with a Rust backend and a Next.js 14 frontend. There is no separate server process.
frontend/
src/ Next.js UI (React 18, Tailwind, BlockNote editor)
src-tauri/ Rust backend (audio capture, Whisper FFI, Pyannote, storage)
whisper-server-package/
Prebuilt Whisper binary used by the Rust audio pipeline
public/ Static assets
- Audio capture: CoreAudio tap on macOS for system audio, plus AVAudioEngine for microphone input. Mixed in Rust with clipping prevention.
- Transcription:
whisper-rswith Metal acceleration. Streaming transcripts delivered through Tauri events. - Diarization: ONNX Pyannote embedding model, clustered in process.
- Summarization: Pluggable provider abstraction in
src-tauri/src/summary/. Each provider implements a commonSummarizertrait.
cd frontend
pnpm install
pnpm run tauri:dev # dev server + Tauri window, hot reload
pnpm run lint # ESLint
pnpm run typecheck # tsc --noEmit
pnpm run tauri:build # release .app + .dmgFrontend dev server: http://localhost:3118.
Pull requests are welcome. Keep the scope tight and the commit history linear. Every PR should:
- Leave
pnpm typecheckclean. - Match the existing visual system. See
DESIGN.mdand the Tailwind config. - Include a one-line changelog entry in the PR description.
Ultra is a fork of the open-source Meetily project (MIT, Zackriya Solutions). Ultra is an independent rebrand with substantial changes to UI, audio pipeline, AI provider layer, and packaging. The original MIT license is preserved.
MIT. See LICENSE.md.
- Brand consistency polish: replaced system blue accent with brand purple across all UI primitives
- Fixed tab gradient, wordmark sizing, About page double scrollbar, and duplicate Language button
- Full changelog in the release notes
- Complete brand refresh to Ultra identity with new logo, color palette, and typography
- New DESIGN.md documenting the complete visual system
- Rebranded across the entire application
- Fixed app launch on macOS 26 and DMG build pipeline
- macOS release packaging and code signing improvements
- Language settings accessible everywhere, sidebar improvements, auto-summary race condition fix
Made by Ashref.