Install
You run this on a computer you leave turned on. It is not a website that holds your audio. There is no installer. Starting with 0.1.11, the npm package ships compiled JavaScript and the inbox UI. Earlier releases through 0.1.10 do not start from npm installations. The selected transcription engine and its Python runtime remain separate prerequisites.
Before you clone
Exercised path: Windows, Node 22.13+, pnpm, and an NVIDIA GPU with CUDA. The checkout scripts use Node's TypeScript support. macOS and Linux remain unverified.
Choose the transcription engine:
- Bark Pro (English, default): isolated Python 3.11 / CUDA 12.8 / BF16 runtime. Set it up with the repository's PowerShell script below. CPU mode is available but unverified.
- Whisper: Python with faster-whisper. Set
transcription.modeltolarge-v3orturboand pointwhisper.pythonat that interpreter. Whisper CPU mode is slower.
Both paths require ffmpeg and ffprobe for audio assessment and preparation, including unfiltered mode. Cleanup through ollanet is optional; without it, raw transcripts remain available.
Get it running
git clone https://github.com/Catalyst-Forge-LLC/dictawhisper.git
cd dictawhisper
cp config.example.json config.json
pnpm install
# Set up the default Bark Pro engine in PowerShell, using your Python 3.11 path:
# & ./scripts/setup-bark.ps1 -Python C:/path/to/Python311/python.exe
# For Whisper instead, configure transcription.model and whisper.python.
pnpm run doctor
pnpm dev
Ready signal: pnpm run doctor exits 0, or prints only warnings (CPU mode, ollanet unreachable, port already in use). Then open http://localhost:7777. Record, or drag a file onto the page. Once the engines are configured, change the default in Settings → Transcription.
pnpm run doctor is the checkup. pnpm doctor without run is pnpm's own command and will not check this app.
If cleanup cannot be reached, that is a warning. Notes still get written. See Install docs for the probe list.
Optional: sync a phone folder
If you already record on your phone, install Syncthing (or any folder sync tool) on both devices and share a folder to your workstation.
Add that directory to "roots" in config.json:
"watch": {
"roots": ["C:\\Users\\YOU\\VoiceNotes"]
}
Files arriving via folder sync wait 30 minutes after the last write before organizing into YYYY/MM/ and transcribing, so recordings never process mid-transfer. This is external folder synchronization, not a DictaWhisper service.
Optional: read from your phone on Tailscale
If you want to open the inbox from a phone or laptop away from your desk, install Tailscale on both devices.
Set "http": { "tailscale": true } in config.json (or DICTA_TAILSCALE=1) and start the app. The checkup and startup logs print a link on this machine's Tailscale address. Open that from a device on the same tailnet. The API stays on 127.0.0.1. The inbox is not published on your LAN. There is no login on the app.
Already have a pile of recordings?
pnpm retranscribe
pnpm retranscribe --dir="./notes/2026/08" --limit=5 --reclean
That walks existing notes, newest first. Flags are in the docs.
Optional: agents, read-only
pnpm mcp is a stdio MCP server over the notes files. Search, fetch a note, list tags, list recent. It does not write. See MCP.