AutoDoc

Open-source meeting notes for Mac and Windows that also record your screen, so you can click a note and replay that moment.

AutoDoc is an open-source, local-first meeting notes app for macOS and Windows. Record your screen and meeting audio, get editable notes, and click a note's timestamp to replay the moment. Transcription, notes, and Ask AI run on your computer. No AutoDoc account or AI API key required.

---

New in v1.3.3: Dark Mode and UI localization in 27 languages, plus bug and security fixes. Read the latest release notes.

---

Features

AutoDoc keeps the whole meeting workflow on your computer, from detection to replay, and every feature below is included, not a paid upgrade.

• On-device transcription. Models are chosen for your meeting language and hardware, with no network round-trips for your audio.
• Local notes, no accounts. Notes run through a local Ollama instance AutoDoc manages. No AutoDoc account, no API keys, no inference bills.
• Multi-track capture. Screen, microphone, and system audio are recorded as separate streams. Your microphone is labeled Me and combined system audio is labeled Them; AutoDoc does not identify individual remote participants.
• Replay key moments. Watch the recording in the Transcript tab, or click a note's timestamp to jump to that moment in the video.
• Editable notes. Get a summary and conversation topics with supporting details, edit them, copy as plain text, or export to PDF, Word, or Markdown. Review generated notes before relying on them.
• Ask AI across meetings. Get grounded answers drawn from your meetings, entirely on-device.
• Calendars and detection. Connect Google and Microsoft calendars with Off / Once / Series auto-record per event. AutoDoc lives in your macOS menu bar or Windows tray and offers to record when a Zoom, Meet, Teams, Webex, or Slack meeting starts.
• Full-text search. Search every transcript and note, with deep links straight to the moment.
• 27 meeting languages on supported hardware. See System requirements for what each machine supports.
• Encrypted at rest. AES-256-GCM protects recordings, transcripts, notes, and metadata. Electron safeStorage protects the key through macOS Keychain or Windows DPAPI when available; otherwise the key is stored locally without operating-system protection. Exported files are separate from this encrypted storage.

See PRODUCT.md for a deep technical breakdown of every subsystem.

Features in action

Download

Platform support: AutoDoc supports macOS 14+ on Apple Silicon and Windows 10+ on x64 PCs. Intel Macs and Windows on ARM are not currently supported.

Ready-to-use installers are available for Mac and Windows. You do not need to build from source.

macOS

1. On the latest release, download the macOS .dmg asset.
2. Open the .dmg and drag AutoDoc into your Applications folder.
3. Launch AutoDoc from Applications. On first run it walks you through granting Screen Recording and Microphone permissions; macOS verifies System Audio Capture separately when recording starts. AutoDoc then downloads its local transcription and AI models (~10 GB, one time).

Windows

1. On the latest release, download the Windows .exe installer asset.
2. Open the installer and follow the setup prompts.
3. Launch AutoDoc. On first run it walks you through capture permissions and downloads the local transcription and Ollama models for your selected meeting language.

That's it — no account, no API keys, nothing to configure. Meeting processing runs locally from here on.

System requirements

Local-first meeting apps need real hardware headroom. AutoDoc adjusts its processing profile to the available hardware while keeping meeting content on your device.

	macOS	Windows
OS version	macOS 14.0 (Sonoma) or later	Windows 10 or later, 64-bit
Processor / chip	Apple Silicon required (M1, M2, M3, M4, or later)	x64 Intel or AMD; 8+ logical processors recommended for CPU-only processing
Memory	8 GB minimum (supported) · 16 GB+ recommended	8 GB minimum (supported) · 16 GB+ recommended
GPU	Built into Apple Silicon; nothing extra needed	Optional. A DirectML GPU with 4 GB+ VRAM accelerates Parakeet transcription; CPU fallback is automatic
Storage	~10 GB free for first-run downloads, plus space for encrypted recordings	~10 GB free for first-run downloads, plus space for encrypted recordings
Network	Needed for the first-run model download; offline afterwards for recording, transcription, and notes	Needed for the first-run model download; offline afterwards for recording, transcription, and notes
Permissions	Screen Recording, Microphone, System Audio Capture	Screen capture, Microphone, System Audio

Transcription on Mac is built on MLX and requires Apple Silicon; there is no Intel or Rosetta fallback.

Lower-spec computers. 8 GB Macs and 8 GB Windows PCs (including ones with a GPU) are English-only and use the smaller llama3.2:3b notes model with sequential audio processing and notes after transcription, so expect a longer wait after the meeting. Windows CPU-only machines with fewer than 8 logical processors or under 16 GB of RAM get the same English-only, lower-impact profile. A GPU does not replace RAM headroom. Settings shows which languages are available on your computer, and the meeting language list covers language-specific requirements.

Build from source

AutoDoc is an Electron + electron-vite app. To build it yourself:

Prerequisites

• macOS: Apple Silicon Mac (M1 or later) running macOS 14+, plus Xcode command-line tools and Homebrew
• Windows: Windows 10+ on x64, plus the prerequisites documented by the setup scripts
• Node.js 20+

Steps

git clone https://github.com/DuetDisplay/AutoDoc.git
cd AutoDoc
npm ci
cp .env.example .env   # optional: configure self-hosting knobs
npm run build:mac      # macOS: produces a DMG under dist/
# or
npm run build:win      # Windows: produces an installer under dist/

To run in development:

npm run dev

If you want calendar integration or your own hosted services in a fork build, see docs/SELF_HOSTING.md for the required environment variables and OAuth setup. Forks do not use Duet's hosted infrastructure by default.

Under the hood

AutoDoc is a single Electron app. The main process owns recording, the transcription and summarization pipeline, encryption, calendar sync, and the local Ollama lifecycle. The renderer is a React UI. All heavy processing happens locally.

	macOS (Apple Silicon)	Windows
English transcription	MLX Whisper distil-large-v3	Parakeet TDT 0.6B v3, DirectML GPU or CPU-optimized
Other languages	Canary or Whisper large-v3-turbo	Canary or Whisper large-v3-turbo
Notes and Ask AI answers	qwen3:4b-instruct by default; llama3.2:3b on lower-spec machines	qwen3:4b-instruct by default; llama3.2:3b on lower-spec machines
Ask AI search embeddings	qwen3-embedding:0.6b	qwen3-embedding:0.6b

flowchart TB
  subgraph app["AutoDoc on your computer"]
    UI["Renderer · React UI"] <-->|IPC| Main["Main process"]

    subgraph pipeline["Local processing pipeline"]
      direction LR
      Rec["Recording<br/>screen · mic · system"]
      Platform{"Current platform"}
      MacTrans["macOS transcription<br/>MLX Whisper"]
      WinTrans["Windows transcription<br/>Parakeet · GPU or CPU"]
      Labels["Audio source labels<br/>Me vs Them"]
      Notes["AI notes<br/>Ollama"]
    end

    Main --> Rec
    Rec --> Platform
    Platform -->|macOS| MacTrans
    Platform -->|Windows| WinTrans
    MacTrans --> Labels
    WinTrans --> Labels
    Labels --> Notes
    Notes --> Store["Encrypted store<br/>AES-256-GCM"]

    Main -.->|"optional calendar"| Cal["Google / Microsoft<br/>Calendar APIs"]
    Main -.->|"optional OAuth token exchange"| Auth["Calendar auth worker"]
  end

Privacy

AutoDoc processes meeting content on-device. Audio, transcripts, and notes are not uploaded to AutoDoc or a cloud AI API. Analytics and crash reporting are strictly opt-in; before opt-in, analytics facts stay local only. Read the full Privacy Policy for exactly what is and isn't collected, and PRODUCT.md for the technical details.

When AutoDoc uses the network

• Required once: the first-run download of local transcription and Ollama models.
• Optional: app updates and additional model downloads. Official public updates can use a Cloudflare mirror with daily aggregate request totals; see docs/UPDATE_MIRROR.md.
• Optional Google/Microsoft calendar sync, where the app talks directly to their APIs. Duet's stateless OAuth worker only exchanges tokens and never receives recordings, transcripts, or notes.
• Optional analytics and crash reporting, only if you opt in.

FAQ

Is AutoDoc a Granola alternative? Yes — a free, open-source one that processes meetings on your computer and keeps a screen recording you can replay. See how AutoDoc compares.

How do I know AutoDoc is recording? While recording, AutoDoc shows a Recording banner in the app (with a timer and stop control) and switches the menu bar / tray icon to a recording state. Meeting detection only offers to start recording; it will not start silently unless you previously enabled calendar auto-record for that event (Once or Series). In that case, recording can begin without another prompt.

What about other participants? AutoDoc records on your computer. It does not join the call as a bot or announce itself to others. You are responsible for following the recording laws and norms that apply to your meeting.

Does AutoDoc include action items in its notes? Action items are currently omitted from generated notes while Next Steps is disabled. Information classified as an action item can also be omitted. Review the recording and transcript when checking follow-up commitments. See the latest release notes for known issues.

Will my existing notes change after updating? Updating does not automatically regenerate your existing meeting notes. You can reprocess a meeting when you want to generate new notes. Older notes remain supported.

Contributing

Contributions are welcome! Please read CONTRIBUTING.md, the ROADMAP.md, and our CODE_OF_CONDUCT.md before opening a pull request. Security issues should follow the process in SECURITY.md.

License

AutoDoc is licensed under the GNU Affero General Public License v3.0 (LICENSE). You're free to use, study, modify, and share it, and if you run a modified version as a network service you must make your source available to its users under the same license.

Acknowledgements

AutoDoc stands on the shoulders of excellent open-source work:

• whisper.cpp — on-device speech-to-text
• NVIDIA NeMo Parakeet — Windows speech-to-text
• ONNX Runtime and DirectML — Windows inference
• faster-whisper and CTranslate2 — compatibility transcription
• Ollama — local LLM runtime
• Apple MLX — Apple Silicon acceleration
• FFmpeg — audio/video processing
• Electron + electron-vite — desktop app foundation

---