VxRuby

A self-hosted personal speech transcriber. Open the browser on your phone, speak into it, and the server transcribes the audio using RubyLLM and copies the text to your macOS clipboard.

Architecture

Phone Browser  →(WebSocket)→  Ruby Server (macOS)  →(audio)→  LLM (RubyLLM)
               ←(WebSocket)←                       ←(text)←
                                      ↓
                                macOS clipboard (pbcopy)

Setup

Prerequisites

• Ruby 3.4+
• A Gemini API key (or any LLM provider supported by RubyLLM)

Install

bundle install

Configure

cp .env.example .env

Edit .env and add your API key:

GEMINI_API_KEY=your_key_here
TRANSCRIPTION_MODEL=gemini-2.5-flash
PORT=4567
BIND=0.0.0.0
PERMITTED_HOSTS=your-hostname.ts.net

localhost is always permitted. Add additional hostnames (comma-separated) via PERMITTED_HOSTS.

SSL (required for mobile microphone access)

Browsers require HTTPS for microphone access from non-localhost origins. Generate a self-signed cert:

mkdir -p certs
openssl req -x509 -newkey rsa:2048 \
  -keyout certs/server.key -out certs/server.crt \
  -days 365 -nodes \
  -subj "/CN=your-hostname" \
  -addext "subjectAltName=DNS:your-hostname,IP:your-ip"

Replace your-hostname and your-ip with your server's hostname and IP address (e.g. Tailscale hostname and IP).

If no certs are present, the server falls back to HTTP (works for localhost testing).

Run

./bin/server

Open https://your-server:4567 on your phone. Accept the self-signed cert warning, then tap Record and speak.

How it works

1. The browser captures audio using the MediaRecorder API
2. Audio chunks are streamed to the server via WebSocket
3. When you tap Stop, the server sends the audio to the configured LLM for transcription
4. The transcribed text is copied to the macOS clipboard and displayed in the browser

License

MIT