AutoByteus Voice Runtime

This project packages the downloadable Voice Input runtime used by the desktop app.

Responsibilities

• pin the upstream whisper.cpp source revision
• build platform-specific whisper-cli binaries
• fetch the selected model asset for release
• generate the runtime manifest consumed by autobyteus-web
• publish versioned runtime assets under this repository's own v* release lane

Intended Output

Each runtime release should publish:

• one binary/runtime asset per supported platform and architecture
• one model asset for the selected v1 model
• one voice-input-runtime-manifest.json file with checksums, URLs, and entrypoints

Suggested Directory Shape

autobyteus-voice-runtime/
  .github/workflows/
  scripts/
  dist/
  metadata/

Local Scripts

• scripts/build-runtime.sh <platform> <arch>
  • downloads the pinned whisper.cpp source tarball
  • builds a static whisper-cli
  • writes the normalized runtime binary into dist/
• scripts/download-model.mjs
  • downloads the configured model asset into dist/
• scripts/generate-manifest.mjs
  • computes SHA-256 checksums
  • emits the final manifest with GitHub release asset URLs

Current Status

This repository owns the app-facing runtime release contract. The next hardening steps are automated validation and end-to-end proof that the desktop app can install, discover, and invoke the published runtime artifacts from this separate release surface.