@swairshah/pi-talk
v1.3.2
Published
Text-to-speech extension for Pi coding agent using Loqui
Maintainers
Readme
pi-talk
Text-to-speech extension for Pi coding agent. Gives Pi a voice using <voice> tags.

Features
- Local TTS - No cloud APIs, runs entirely on your Mac
- Multiple voices - 7 different voice options
- Central playback - Pi sends speech jobs to Loqui's local broker queue
- Configurable - Toggle verbosity, mute, change voices
Requirements
Loqui.app must be installed and running (provides the TTS server + local broker queue).
brew install swairshah/tap/loquiThen launch Loqui from Applications - it runs in the menubar.
Installation
pi install npm:@swairshah/pi-talkUsage
Once installed, Pi will automatically speak <voice> tagged content in its responses.
Commands
| Command | Description |
|---------|-------------|
| /tts | Toggle TTS on/off |
| /tts-mute | Mute audio (keeps voice tags in responses) |
| /tts-voice <name> | Change voice (alba, marius, javert, fantine, cosette, eponine, azelma) |
| /tts-style | Toggle between succinct and verbose voice prompts |
| /tts-say <text> | Speak arbitrary text |
| /tts-stop | Stop current speech |
| /tts-status | Show current status |
Global Shortcut
Press Cmd+. to stop speech at any time (requires Loqui.app running).
How it works
- The extension injects a system prompt that teaches Pi to use
<voice>tags - When Pi responds, the extension extracts
<voice>content - The extension sends speech jobs to Loqui's local broker (
127.0.0.1:18081), includingsourceApp,sessionId, andpid - Loqui schedules per-session queues and plays audio centrally (no cloud, no API keys)
Credits
- TTS model: Pocket TTS by Kyutai Labs
- Rust implementation: pocket-tts by babybirdprd
License
MIT
