Two streams, captured separately
System audio and your microphone are recorded as independent tracks via cpal and ScreenCaptureKit, so the transcript stays clean and attributable.
Folio captures system audio and your microphone, transcribes on-device, and writes a clean markdown note per meeting to your own vault. Private by default.
brew install --cask foliotitle: Launch sync
attendees: You, Speaker 2, Speaker 3
duration: 28m · model: whisper-large-v3
source: on-device
Let's lock the launch date before we scope the rest.
Works for me. I'll own the migration checklist.
I can have the pricing page ready by Thursday.
Decision: ship on the 24th, freeze scope Friday.
On-device transcription
Metal-accelerated Whisper
Works fully offline
Privacy Mode airgap
No bots in your calls
System audio capture
Open source
Apache-2.0 licensed
Folio sits in the menu bar, watches your calendar and audio devices, and turns each conversation into a note you actually own.
System audio and your microphone are recorded as independent tracks via cpal and ScreenCaptureKit, so the transcript stays clean and attributable.
The default path keeps audio, transcripts, and notes on your machine. Privacy Mode physically blocks every outbound call except localhost.
Speakers are separated locally with pyannote segmentation and a speaker-embedding model. Your microphone is always labelled You.
Every meeting becomes a portable markdown file in your own vault, with frontmatter for attendees, duration, model, and source.
A local MCP server gives Claude, Cursor, and Claude Code read-only access to your transcripts, tasks, and memory over stdio.
Plain files on disk, an open IPC contract, and an Apache-2.0 codebase. No lock-in, no account, no cloud dependency.
Four steps, all on your machine. The only time anything leaves is the one-time model download or an opt-in cloud call you trigger yourself.
Folio watches your calendar and audio devices through EventKit. No Google OAuth, no Microsoft Graph, no meeting bots.
When the call starts, system audio and your microphone are captured as separate streams and written to WAV on disk.
The bundled Whisper backend transcribes on-device and diarizes speakers. Cloud transcription is an explicit opt-in, never the default.
A markdown file lands in your vault with frontmatter, speaker-labelled transcript, decisions, and tasks. Searchable forever.
Folio is built so the private path is the easy path. You can run an entire meeting end-to-end with Wi-Fi off.
Full details, retention, and the recording-consent guide live in the privacy documentation. Folio gives you the tool; obtaining consent is your responsibility.
{
"mcpServers": {
"folio": {
"command": "folio-mcp",
"args": []
}
}
}Folio ships a local MCP server. Any MCP-aware tool gets read-only access to your transcripts, tasks, and memory over stdio — no cloud, no proxy.
Requires macOS 13+ on Apple Silicon or Intel. The first run downloads the model weights once, then everything is local.
brew tap woosal1337/folio https://github.com/woosal1337/folio
brew install --cask folioAlready installed? Update any time with brew upgrade --cask folio.
Install Folio, grant microphone and screen-recording permission, and your next meeting writes itself to your vault.
brew install --cask folio