opencode-tts

An OpenCode plugin that automatically speaks assistant responses when a session goes idle.

https://github.com/user-attachments/assets/cf1c3166-dbd7-43b8-8215-1ea12d336463

How it works

The plugin listens for session idle events and automatically:

1. Captures the latest assistant message
2. Either summarizes it (default) or uses the full text (configurable)
3. Converts it to speech using a local TTS engine

Install

Tell OpenCode:

Install plugin opencode-tts

Or add to your OpenCode config (~/.config/opencode/opencode.json):

{
  "plugin": ["opencode-tts"]
}

Modes

• summary (default): Auto-speaks a summarized version of each response
• full: Auto-speaks the complete response text

User controls

Use these slash commands to manage TTS behavior:

| Command | Effect | | --- | --- | | /tts-mode-summary | Switch to summary mode (default) | | /tts-mode-full | Switch to full text mode | | /tts-on | Enable automatic TTS | | /tts-off | Disable automatic TTS | | /tts-speak <text> | Speak arbitrary text immediately | | /tts-repeat | Re-speak the last response in the current session | | /tts-uninstall | Remove plugin files and clean up opencode.json |

Mode and status changes persist across sessions.

Configuration

Plugin settings live in ~/.config/opencode/plugins/opencode-tts.jsonc:

{
  "enabled": true,          // false to disable TTS — persisted across sessions
  "mode": "summary",        // "summary" | "full" — persisted across sessions
  "debug": false,
  "backend": "edge_tts",    // "edge_tts" | "say"
  "voice": "...",           // voice override for the active backend
  "summaryLength": "2 sentences", // free-text length hint passed to the LLM (e.g. "30 words", "3 short sentences")
  "summaryProvider": "...", // provider ID for the summary model
  "summaryModel": "...",    // model ID for the summary model
  "edge_tts": {
    "voice": "en-US-AvaNeural",
    "rate": "+25%",
    "volume": "+0%"
  }
}

The plugin auto-installs edge-tts into a managed Python venv at ~/.config/opencode/tts-venv on first use.

Summary model

The plugin resolves the model for summary generation in this order:

1. summaryProvider + summaryModel from the plugin config
2. small_model from your OpenCode config
3. The same model that produced the assistant response
4. Text fallback: first 40 words of the raw response

TTS backends

edge_tts (default): Uses Microsoft Edge's neural TTS. Auto-installed. Requires an audio player: afplay (macOS built-in), ffplay, or mpg123.

The default voice is en-US-AvaNeural, which is recommended for English. Edge TTS supports a wide range of languages and locales — you can use any multilingual voice (e.g. en-US-AvaMultilingualNeural) or switch to a completely different language (e.g. it-IT-ElsaNeural for Italian, fr-FR-DeniseNeural for French). Run edge-tts --list-voices to see all available voices.

say: Uses macOS say or Linux spd-say/espeak. No install needed.

To use a custom edge-tts executable:

{
  "edge_tts": {
    "command": ["python3", "-m", "edge_tts"]
  }
}

Local development

Build:

npm install
npm run build

Checkout and point to local plugin:

{
  "plugin": ["file:///path/to/opencode-tts/dist/index.js"]
}