Private, local-first call & meeting recording for macOS. Capture system audio from any call (Google Meet, Zoom, Teams, or anything that plays through your speakers), transcribe it, separate who said what, and turn it into clean Markdown notes — without a bot joining your meeting and without your audio leaving your machine unless you choose a cloud transcription engine.
A local alternative to Fathom/Otter-style meeting recorders. No meeting bot, no mandatory cloud, no "who invited the notetaker?" — it captures the audio already playing on your Mac.
Hosted meeting recorders join your call as a visible bot and upload everything to their servers. CallCapture instead taps the audio your Mac is already producing (via Core Audio process taps, macOS 14.2+), so:
- No bot appears in the meeting.
- Local by default — on-device Whisper transcription means audio never leaves your machine. Cloud engines are opt-in.
- You own the data — recordings, transcripts, notes, and the session database all live in your local Application Support folder.
- 🎙️ System-audio capture — records call audio via Core Audio process taps; optional microphone capture mixed in.
- 🧑🤝🧑 Speaker separation — captures your mic and the remote side as separate stems and attributes turns with on-device diarization (FluidAudio), so transcripts read You: / Speaker 1: instead of one wall of text.
- 📝 Transcription — on-device Whisper (
pywhispercpp) or cloud engines (AssemblyAI, Deepgram Nova-3) with automatic per-language routing. - 🌍 Multi-language — broad language support with per-recording spoken language and a separate output/notes language.
- 🧠 Conversation insights — sentiment, acoustic emotion, key points, and action items via an LLM post-processing pass (OpenRouter or a local model).
- 🗒️ Markdown notes — meeting-notes / transcript / Obsidian export profiles, with one-click "Save to Vault."
- 🪶 Menu-bar app — lightweight, stays out of the way.
Session detail — transcript with speaker attribution (You / Speaker 1), the generated Markdown note, and conversation insights (sentiment per speaker, emotional arc, recommended actions, action items):
Settings — pick a transcription engine (on-device Whisper or cloud), per-language provider routing, LLM post-processing, on-device diarization & emotion models, and export to an Obsidian vault:
CallCapture is two cooperating pieces:
| Component | Tech | Role |
|---|---|---|
macOS app (macos-app/) |
Swift, SwiftUI, Core Audio, GRDB | Menu-bar UI, audio capture, on-device diarization, session database, settings |
Python worker (python-worker/) |
Python 3.11+, Pydantic, Click | Transcription + analysis pipeline; invoked per job over stdin/stdout JSON |
The app spawns the worker as a subprocess, sends a JobRequest on stdin, and reads progress + a JobResult back. See docs/DEVELOPMENT.md for the full data flow and build instructions.
- macOS 14.2 or later (Core Audio process taps)
- Apple Silicon recommended
- Python 3.11+ (for the worker)
- API keys only if you opt into cloud engines (AssemblyAI / Deepgram for transcription, OpenRouter for LLM notes) — stored in the macOS Keychain, never in the repo.
brew tap bodharma/callcapture
brew install --cask callcapture
# update to the latest release later:
brew upgrade --cask callcaptureNotarized by Apple — it opens with a normal double-click, no Gatekeeper bypass.
The packaged app uses cloud transcription (AssemblyAI / Deepgram) plus full on-device analysis (diarization, sentiment, emotion, insights). On-device Whisper is available in source builds only for now — see docs/DEVELOPMENT.md.
git clone https://github.com/bodharma/callcapture.git
cd callcapture
# Python worker
cd python-worker
python3 -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
cd ..
# Build + launch the app in dev mode (builds Swift, bundles, launches)
./run-dev.shA waveform icon appears in the menu bar. Start a call, hit record, stop, and let it transcribe. Configure engine, languages, API keys, and export folders in Settings.
- On-device Whisper keeps audio fully local.
- Cloud engines (AssemblyAI, Deepgram, OpenRouter) upload audio/transcript only when you select them.
- API keys live in the macOS Keychain. Recordings and the session DB stay in
~/Library/Application Support/CallCapture/and are git-ignored.
GNU AGPL-3.0 © bodharma. If you run a modified version as a network service, you must release your source.
Issues and PRs welcome. Start with docs/DEVELOPMENT.md for architecture, build, and test workflows.