OpenCode Voice Plugin

Voice input plugin for OpenCode with PyQt5 stop button and ntfy.sh notifications.

Features

• 🎙 Voice Recording - Record audio with a floating PyQt5 stop button
• 📍 Smart Positioning - Button appears at bottom-right of Terminal window
• 🔔 ntfy.sh Notifications - Get notified when to speak, when transcribing, and the result
• 🇫🇷 French Whisper - Uses medium model for accurate French transcription
• ✅ Visual Feedback - Button changes color: 🔴 → ⏳ → ✅/❌

Commands

Command	Description
/talk	Record voice, transcribe, and auto-send as message
/speak	Ask AI to speak in French with elia-speak

Installation

1. Prerequisites

macOS requires:

• sox (for audio recording)
• Python 3.10+ with pip
• OpenCode CLI

brew install sox

Python packages:

pip install -r requirements.txt

Or manually:

pip install openai-whisper PyQt5

2. Install Commands

mkdir -p ~/.config/opencode/commands

Copy these files to ~/.config/opencode/commands/:

• talk.md - Voice recording command
• speak.md - French summary command
• voice_record.py - Main Python script

cp talk.md speak.md voice_record.py ~/.config/opencode/commands/
chmod +x ~/.config/opencode/commands/voice_record.py

3. ntfy.sh Setup

The plugin uses ntfy.sh/OpenCode topic for notifications.

Option A: Use default (public) No setup needed. Install ntfy app and subscribe to OpenCode topic.

Option B: Use private topic Edit voice_record.py and change:

def notify(msg):
    subprocess.run(
        ["curl", "-s", "-X", "POST", "https://ntfy.sh/YOUR_TOPIC", "-d", msg],
        capture_output=True,
    )

4. Download Whisper Model

The first run will download the medium model (~1.5GB). To pre-download:

python3 -c "import whisper; whisper.load_model('medium')"

Usage

/talk - Voice Message

1. Type /talk in OpenCode
2. A 🔴 red stop button appears at bottom-right of Terminal
3. Speak your message
4. Click ⏹ Stop (or wait 60s timeout)
5. Button turns ⏳ blue during transcription
6. Button turns ✅ green with notification containing transcript
7. Transcript is auto-inserted and sent as chat message

/speak - French Summary

Type /speak to ask the AI to speak in French using elia-speak with Kokoro TTS. The AI will prefix the message with "Speak about the :" followed by the content to be spoken.

How It Works

┌─────────────────────────────────────────────────────┐
│  User types /talk                                   │
│         ↓                                           │
│  voice_record.py launches                           │
│         ↓                                           │
│  🔴 Red button appears (bottom-right of Terminal)   │
│         ↓                                           │
│  ntfy notification: "🎙 Parle maintenant..."         │
│         ↓                                           │
│  User clicks ⏹ Stop                                 │
│         ↓                                           │
│  ⏳ Blue button (transcribing...)                    │
│         ↓                                           │
│  ntfy notification: "⏳ Transcription..."            │
│         ↓                                           │
│  Whisper (medium model) transcribes in French        │
│         ↓                                           │
│  ✅ Green button                                    │
│         ↓                                           │
│  ntfy notification: "✅ [transcript]"              │
│         ↓                                           │
│  Transcript printed → OpenCode sends as message     │
└─────────────────────────────────────────────────────┘

File Structure

~/.config/opencode/commands/
├── talk.md           # Command: record voice
├── speak.md          # Command: French summary
└── voice_record.py   # Main Python script

~/.config/opencode/plugins/
└── voice.ts          # (Optional) Plugin alternative

Troubleshooting

"rec command not found"

brew install sox

"No module named 'whisper'"

pip install openai-whisper

"No module named 'PyQt5'"

pip install PyQt5

Button not appearing at Terminal

The script uses AppleScript to detect Terminal/iTerm2 window position. Grant Accessibility permissions if needed:

• System Preferences → Security & Privacy → Privacy → Accessibility
• Add Terminal/iTerm

Notifications not working

Check your ntfy.sh topic. Default is OpenCode. Install the ntfy app and subscribe to your topic.

Whisper slow on first run

First run downloads the model (~1.5GB). Subsequent runs use cached model.

Configuration

Edit voice_record.py to customize:

TMP_FILE = "/tmp/opencode_voice.wav"  # Temp audio file
WHISPER_MODEL = "medium"              # Model: tiny/base/small/medium/large
WHISPER_LANG = "fr"                   # Language code
NTFY_TOPIC = "OpenCode"               # ntfy.sh topic

Requirements

• macOS (uses AppleScript for window detection)
• sox (brew install sox)
• Python 3.10+
• openai-whisper
• PyQt5