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. Python and faster-whisper remain separate prerequisites.

Before you clone

Exercised path: Windows, Node 22.13+ (the scripts use --experimental-strip-types), pnpm, a Python interpreter with faster-whisper, and an NVIDIA GPU with CUDA (whisper.device = cuda). Point whisper.python at that interpreter.

Also required for the default denoise path: ffmpeg on PATH.

Optional: ollanet plus an Ollama model for text cleanup (this computer or another you already use). Without it you still get the raw transcript.

CPU mode (whisper.device = cpu) is supported and marked slow. macOS and Linux have not been recorded as verified in this project. Treat them as unverified.

Get it running

git clone https://github.com/Catalyst-Forge-LLC/dictawhisper.git
cd dictawhisper
cp config.example.json config.json
# set whisper.python, whisper.device, and (optionally) ollanet.machine / cleanModel
pnpm install
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.

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.