Skip to content

openclaw-full-ml

Version 2026.144.1443
Repo superproject

OpenClaw full headless stack plus the CUDA speech-ML tools (whisper STT + sherpa-onnx TTS) openclaw-full-ml is a metalayer that installs nothing of its own and composes three candies, so the observable effect is the union of their key artifacts:

  • openclaw-full — the OpenClaw gateway plus every headless CLI tool (ffmpeg, ripgrep, gh, tmux, sqlite, uv, …). The gateway binary lands at ~/.npm-global/bin/openclaw and ffmpeg/rg at /usr/bin.
  • whisper — OpenAI Whisper speech-to-text installed via pixi into the shared default environment, exposing the whisper console script at ~/.pixi/envs/default/bin/whisper (ffmpeg is its audio decoder).
  • sherpa-onnx — offline text-to-speech whose VITS piper voice model is downloaded and extracted to ~/.local/share/sherpa-onnx/models/vits-piper-en_US-lessac-high. Each branch is verifiable by the presence of its landed artifact, so a no-op composition fails every check below.

This candy’s plan: — the runnable spec charly check executes against a live deployment. check: steps are idempotent probes; run: steps change state.

Intent Step
check the OpenClaw gateway binary is installed (provided by openclaw-full)
check ffmpeg (whisper’s audio decoder, from openclaw-full) is installed at /usr/bin/ffmpeg
check the ripgrep search binary (from openclaw-full) is installed at /usr/bin/rg
check the whisper speech-to-text console script is installed in the default pixi environment
check the sherpa-onnx offline TTS voice-model directory is extracted under the user data dir
agent-check the running OpenClaw gateway can transcribe audio through whisper and synthesize speech through sherpa-onnx end to end