It does one thing: turns text into speech and plays it without holding up the conversation. The persistence is what makes it comfortable in practice — you pick a voice and a tone once and every later summary arrives in it. Synthesis is billed to your own OpenAI key, so cost tracks how much you have read to you.
A text-to-speech output for whatever the assistant just wrote. It sends text to OpenAI's TTS models, plays the result through the platform's own audio player without blocking, and cleans up the temporary file afterwards. Voice and tone are saved between calls, so they are set once rather than passed every time.
- A summary spoken aloud and played in the background
- A voice chosen from ten OpenAI voices — alloy, ash, ballad, coral, echo, fable, nova, onyx, sage and shimmer — and remembered
- Delivery instructions saved as a tone, so how it is read is a preference rather than an argument
Python 3.8 or higher and an OpenAI API key with access to the TTS models. macOS and Windows use their built-in players; on Linux one of mpg123, sox, ffmpeg, vlc or alsa-utils needs to be present.
One command plus a key — pip install -e ., then supply credentials