gptme-tts

Text-to-speech (TTS) plugin for gptme using Kokoro.

Installation

pip install gptme-tts

Usage

Start the TTS server (included in the plugin directory):

# tts_server.py is a uv script with inline dependencies — run directly
./tts_server.py

Then start gptme normally — it will detect the running server and enable TTS automatically.

Environment Variables

• GPTME_TTS_VOICE: Voice to use (depends on server backend, e.g. am_adam)
• GPTME_TTS_SPEED: Playback speed multiplier (default 1.0)
• GPTME_VOICE_FINISH: Set to 1 to wait for speech to finish before exiting

Design Notes

Speak modes

The plugin exposes two ways to trigger speech:

• Hook mode (default): A generation_post hook automatically speaks every assistant message.
• Explicit mode: The speak() tool function lets the agent speak specific text on demand.

Currently both are active simultaneously, which can be redundant. A future version may let users configure one or the other via an env var (e.g. GPTME_TTS_MODE=explicit to disable the hook). If you have a use case that requires one mode over the other, please open an issue.

TTS server

tts_server.py is a standalone uv script with inline dependencies (no separate install needed). In a future version it may be started transparently when TTS is first used, rather than requiring manual startup.