Local text-to-speech voice replies for AI coding agents. Your agent speaks its answers aloud through the Kokoro neural TTS engine running entirely on your machine — no cloud service, no API key, no per-word billing.
Works with any coding agent like Claude Code, Codex, OpenCode, etc.
Tell your agent:
Set up super-speech from github.com/reasonmethis/super-speech
SETUP.md is an agentic playbook — your agent runs it end to end: locate the
scripts, install the Python dependencies, download the Kokoro voice model,
verify audio works, self-heal any failures, and pick a default voice.
First-run download: the Kokoro v1.0 model files total about 338 MB
(kokoro-v1.0.onnx ~311 MB + voices-v1.0.bin ~27 MB). They download once
during setup into ~/.super-speech/models/kokoro/ and are reused afterward.
Once set up, just ask your agent to reply using super-speech, e.g.
Use super-speech for your replies until i tell you otherwise
The super-speech skill handles chunking, the drainer lifecycle, and
voice selection. See skills/super-speech/SKILL.md for the full chunking
contract and TTS details.
drainer-kokoro.py (the TTS engine) and speak.sh are cross-platform.
ensure-drainer.sh — the drainer launcher — is written for Windows + git-bash;
on macOS/Linux the setup playbook tells the agent how to substitute the short
POSIX equivalents. The runtime home ~/.super-speech/ is the same on every OS.
MIT — see LICENSE.