Architecture

Fuse Intelligence products share a local-first philosophy: inference and agent CLIs on-device where possible, licensing and sync via the company web API, and one account for every product. VocalFuse is a native C++ dictation shell; VibeFuse is a Windows AI agent harness with a React widget canvas — both validate keys against the same PHP + MySQL portal.

VibeFuse — widget canvas

VibeFuse splits work between a desktop main process (PTY sessions, CLI/MCP management, local Whisper + Piper speech, marketplace widget + skill sync, notes sync, in-app updates, persistence under C:\vibefuse\) and a renderer React canvas with draggable widgets, xterm terminals, embedded browsers, Files explorer, Vibe Agent chat, Skills rail, voice dock, and a thirteen-tab Settings modal.

VibeFuse architecture: installer and Fuse Intelligence web services connect to the desktop main process with PTY and IPC, which bridges to the React renderer canvas, agent CLIs, local voice engines, MCP servers, and marketplace widgets
VibeFuse: main ↔ renderer bridge, live PTY widgets, local voice pipeline, MCP stdio servers, and marketplace packages.
  • Installer — Verifies VF- key, downloads VibeFuse.exe, caches Whisper + Jarvis Piper voice, installs whisper-cli.exe and piper.exe, and pulls missing agent CLIs
  • Main process — node-pty shells, MCP stdio servers (Google Workspace / Drive presets + custom), marketplace widget + skill sync/publish, notes sync, file explorer IPC, state in C:\vibefuse\vibefuse.json
  • Renderer — Session tabs, free-form canvas layouts, Skills rail, voice dock (hold-to-talk, Jarvis wake word, Notes mode), widget:// iframe marketplace widgets, Vibe Agent skill tools
  • Web services — fuseintelligence.org for licensing, version checks, marketplace catalog, notes API, and the update feed (/updates/vibefuse)

Deep dive: VibeFuse docs → Architecture · Product: /products/vibefuse

VocalFuse — native dictation shell

VocalFuse architecture flow: desktop app performs a license check against the PHP + MySQL API, then runs local transcription on-device

VocalFuse components

  • Desktop shell — Native C++ Windows app with floating pill overlay, system tray, and hotkeys
  • On-device transcription — Local transcription model (~57 MB) stored under C:\VocalFuse\models\, running on GPU (CUDA) or CPU
  • License service — Validates subscription; offline grace period 7 days
  • Web portal — PHP, Bootstrap, Tailwind, MySQL (this project)

VocalFuse docs: /docs/vocalfuse · For canvas-scoped free STT/TTS, see VibeFuse voice (not a substitute for system-wide dictation).