VibeFuse Documentation
Free Windows AI agent harness. This hub covers install and your VF- key.
Every feature now has its own page — Settings, voice, Piper, Whisper, Local Servers, HyperFrames, Extensions, widgets, and selling — written against the live desktop UI (14 tabs, Extensions not Tools, Jarvis on the canvas).
Open the full VibeFuse UI catalog — HTML/CSS clones of the desktop shell, settings, widgets, and overlays.
Section guides
Pick a page. Old /docs/vibefuse#voice style links redirect here automatically.
Register or sign in. VibeFuse is free — a VF- key is written to Product Keys.
Download VibeFuseInstaller.exe. It fetches the app, voice models, and local speech engines, then launches VibeFuse.
Spawn Jarvis, Terminal, or Browser from the left rail. Hold the orb to speak — or say “Jarvis”. Connect a provider in Settings → Agent.
Overview
VibeFuse (VibeFuse.exe) is a Windows 10+ x64 widget canvas. Agents, a system terminal, browsers, files, and Jarvis chat each spawn as draggable widgets. Voice is local-first: Whisper STT, Jarvis wake word, Note-Taking, Transcribe, and Piper replies. Local Servers host Piper, Whisper, HyperFrames Studio, Skill Seekers, and Crawl4AI + SearXNG. The marketplace lets you sell widgets, skills, and styling packs through Stripe Connect (80% seller payout).
- 14 Settings tabs — grouped Workspace / Assistants / Library / System. MCP lives on Extensions.
- Widgets — Terminal, Browser, YouTube, Files, Jarvis, Hermes Agent, Claude, Codex, Gemini, Cursor, Qwen, HyperFrames, Video, Code Studio, saved pages, marketplace iframes.
- Skills —
SKILL.mdpackages inC:\vibefuse\skills. Publish from Settings → Skills. - Styling packs —
C:\vibefuse\styling. Apply and publish from Settings → Appearance. - Marketplace — desktop tab publishes widgets only. Price and listing art are on the website.
App shell — title bar, rail, sessions, canvas, and voice dock
Architecture
The main process owns PTY sessions, CLI and MCP management, local speech, marketplace sync, notes sync, in-app updates, and persistence under C:\vibefuse\.
The renderer is the React canvas: session tabs, widgets, Skills rail, voice orb, and the 14-tab Settings dialog.
Deep dive: Architecture.
Requirements
| Component | Minimum | Recommended |
|---|---|---|
| OS | Windows 10/11 64-bit | Latest Windows 11 |
| CPU / RAM | 4-core · 8 GB | 8+ cores · 16 GB with several agent widgets |
| Disk | ~400 MB for the app | SSD; extra space for Whisper / Piper models and agent CLIs |
| Node.js | Installed by the installer (winget) if missing | Current LTS on PATH |
| ffmpeg | On PATH for local microphone transcription | Recent static build on PATH |
| Network | HTTPS to fuseintelligence.org for download, marketplace, notes, updates | Also useful for remote voice hosts and your chat provider |
| Microphone | Required for hold-to-talk and the wake word | Grant Windows Privacy → Microphone to VibeFuse |
Installer
VibeFuseInstaller.exe is a styled navy / steel / gold bootstrap. It reads the live version, downloads the desktop exe from /downloads/VibeFuse/, then installs anything missing before launch.
- Open Downloads or Account → Downloads.
- Copy your complimentary
VF-key from Account → Product Keys. - Run
VibeFuseInstaller.exe, paste the key, then approve the elevation prompt. - The installer saves
VibeFuse.exeunderProgram Files\VibeFuse(elevated) or%LOCALAPPDATA%\Programs\VibeFuse. - Missing pieces are installed only if absent: Node.js LTS and Git via winget; Claude Code, Codex, Gemini, and Qwen via npm; Cursor via PowerShell.
- Whisper models and the Jarvis Piper voice land in
%LOCALAPPDATA%\VibeFuse\models\;whisper-cli.exeandpiper.exeland in%LOCALAPPDATA%\VibeFuse\bin\. - Desktop and Start Menu shortcuts are created, then the app launches.
- Verify product key
- Read latest version
- Download VibeFuse.exe
- Node.js runtime
- Git
- Claude Code CLI
- Codex CLI
- Gemini CLI
- Cursor CLI
- Qwen CLI
- Whisper model
- Jarvis voice + speech engines
- Shortcuts
- Launch VibeFuse
Installing Claude Code CLI…
Updating from inside the app
When a newer build exists, use Settings → About (Check again / Update). Settings → Updates shows the same feed as /updates/vibefuse.
Version
Current release: v1.1.1.
| Source | Where |
|---|---|
| Database | app_version.product_slug = vibefuse — Admin → App Versions |
| Public JSON | /downloads/VibeFuse/version.json |
| API | /api/app-version.php?product=vibefuse |
| Desktop app | Title bar subtitle and Settings → About |
License key
VibeFuse is complimentary. On register or Google sign-in the site creates a unique VF-xxxx-xxxx-xxxx-xxxx key. Keys do not expire.
Sign in to see the key on this page, or copy it from Product Keys.
The key powers marketplace sync, publishing, and notes sync. The installer writes it beside the exe and in %APPDATA%\VibeFuse\license.json. Confirm it under Settings → About.
Troubleshooting
Feature-specific fixes live on the section pages. Common ones:
| Symptom | Fix |
|---|---|
| Voice does nothing | Allow the microphone. Settings → Voice → Refresh devices. Local STT needs ffmpeg on PATH. Use Settings → Local Servers → Test Piper (it also reports Speech). |
| Wake word not triggering | Enable Settings → Voice → Listen for a wake word. Normal Mode only. Say “Jarvis” clearly. |
| Piper / Speech chip Down | Keep Use local Piper / Use local Whisper on. Re-run the installer so %LOCALAPPDATA%\VibeFuse\bin has piper.exe and whisper-cli.exe. |
| HyperFrames will not start | Settings → Local Servers → Start Studio. Workspace stays under C:\vibefuse\hyperframes. See HyperFrames. |
| Skill Seekers failed | Python 3.11+ and Git. Connect again so the venv rebuilds at C:\vibefuse\Skill_Seekers. |
| Google Workspace not ready | Settings → Extensions — Sign in with Google, Enable Workspace APIs, then Connect. |
| Cannot set a paid price | Price is set on the website, not in the app. Finish Stripe Connect. Minimum $0.50. |
| Widget missing from the rail | manifest.json needs id, name, and version. Restart, or ask Jarvis to open it. See Build a widget. |
| CLI missing from the rail | Only installed agents show. Install from Settings → Agent, then Rescan. |
VibeFuse FAQ
Is VibeFuse free?
Yes. Every Fuse Intelligence account gets a complimentary VF- license at no charge. There is no checkout and no monthly fee for VibeFuse. Create a free account, copy your key from Account → Product Keys, and install from /downloads/VibeFuse/.
What harness tools does VibeFuse include?
VibeFuse is a high-capability Windows AI agent harness: Terminal (PowerShell), Browser, YouTube, Files explorer, a Vibe Agent chat assistant that can create and edit Agent Skills, AI coding agent widgets for Claude Code, Codex CLI, Gemini CLI, Cursor CLI, and Qwen, Google Workspace and Drive MCP connections, and a live marketplace for community widgets and SKILL.md packages.
Does VibeFuse include free voice transcription?
Yes. Hold the title-bar voice dock to record. Speech is transcribed locally on your PC with whisper.cpp for free — useful for dictating prompts into AI coding agents without cloud audio uploads.
Does VibeFuse support a wake word?
Yes. Enable the Jarvis wake word in Settings → Voice and say “Jarvis”. VibeFuse plays a short “Yes, sir?” acknowledgement, then records your command. Local Whisper and Piper keep the whole loop on your PC.
How does Note-Taking mode work?
Switch the voice dock to Note-Taking, click the orb to record, and click again to stop. VibeFuse transcribes the take locally, summarizes it into bullet points with a title, and uploads it to your profile — review notes at Account → Notes or in Settings → Notes.
Can the assistant control the canvas layout?
Yes. Jarvis (voice) and the Vibe Agent chat widget can spawn, tile, resize, rename, focus, and remove widgets on command — including built-in panels and custom marketplace widgets other people made. Example: “Jarvis, open three terminals and tile them.” Layouts use tile/fill/row/column arrange modes or exact x/y/w/h coordinates.
Does VibeFuse include free text-to-speech (TTS)?
Yes. Replies are spoken by Piper with the Jarvis voice, running on-device. With an OpenAI-compatible chat provider configured, the voice assistant speaks a short reply line — free local TTS for developer workflows, not a cloud TTS subscription.
Is VibeFuse open source?
VibeFuse is a free Fuse Intelligence product that wraps and runs popular open-source and vendor AI coding CLIs (Claude Code, Codex, Gemini, Cursor, Qwen) in one desktop canvas. Agent CLIs remain theirs; VibeFuse is the free harness, marketplace, voice, and session layer.
How do I install VibeFuse?
Register for free, copy your VF- key from Account → Product Keys, download VibeFuseInstaller.exe from /downloads/VibeFuse/, paste the key in Setup, and launch. The installer fetches the app and missing local voice models before first run.
Where is the widget marketplace?
Browse fuseintelligence.org/marketplace — save widgets and Agent Skills to your library while signed in. Widgets sync into C:\vibefuse\widgets and skills into C:\vibefuse\skills on next launch.
How do I publish a widget to the marketplace?
Build a folder with manifest.json and index.html, open Settings → Marketplace in the desktop app, and publish the folder. Your listings appear publicly on fuseintelligence.org/marketplace, and you can unpublish or republish any time.
What are Agent Skills in VibeFuse?
Agent Skills are Claude/Codex-compatible SKILL.md folders in C:\vibefuse\skills. They appear in the left-rail Skills section. Save them from the marketplace, drop a folder on disk, or ask Vibe Agent to create_skill / update_skill. Any connected provider agent can read the same files.
Can Vibe Agent create and edit skills?
Yes. Vibe Agent (and Jarvis) can list_skills, create_skill, update_skill, delete_skill, execute_skill, read_skill, and read_skill_file. The Skills rail and Settings → Skills refresh as soon as a skill is written.
How do I connect Google Workspace or Google Drive MCP?
Open Settings → Extensions. Use the Google Workspace card (Gmail, Drive, Calendar, Docs, Sheets, Tasks, Slides, Meet, Contacts) or the Google Drive card (read-only Drive). Sign in with Google once — both cards share the same OAuth account — then Connect. Enable Workspace APIs in Google Cloud if tool calls fail.
Which AI providers does VibeFuse support?
Settings → Agent includes OpenAI, Anthropic, Kimi, Z.ai, Ollama (Local), Ollama Cloud, NVIDIA NIM, Featherless, Google Gemini, X.ai, Together AI, OpenRouter, Plugsky, DeepSeek, and Custom OpenAI-compatible endpoints. Each card keeps its own API key, endpoint, and model.
What is Settings → Local Servers?
The fourteenth Settings tab: Use local Piper + Piper endpoint + Test Piper; Use local Whisper + Whisper endpoint (the model picker is on Settings → Voice); HyperFrames Studio start/stop; Skill Seekers Connect; and Crawl4AI + SearXNG endpoints. Voice, speed, and volume live on Settings → Voice.
Can I sell skills I create in VibeFuse?
Yes. Finish Stripe Connect, set a price on Account → Marketplace, and publish widgets, skills, or styling packs. Founders 50 keep 100% after Stripe processing; standard sellers keep 80%. Buyers get the package in C:\vibefuse\skills, \widgets, or styling folders. Custom widgets remain agent-drivable after install.
Can the agent control custom widgets from the marketplace?
Yes. That is the VibeFuse proof point versus Grok Bot, Perplexity Computer, ChatGPT agent, and Claude Cowork. Marketplace widgets and agent-made UI (manifest.json + index.html) are still widgets: spawn, tile, focus, and operate them by name. Cloud computers give you a vendor desktop, not a public catalog of user-built agent UI.
What is HyperFrames in VibeFuse?
A local HTML-to-video studio on port 4567. Vibe Agent can compose a video about your workspace folder, render an MP4 to C:\vibefuse\hyperframes\renders, and spawn a Video widget.
What is Skill Seekers?
A local Python MCP cloned to C:\vibefuse\Skill_Seekers. It searches docs-to-skill presets (React, Godot, Blender, Claude Code, Unity, and more) before you create a duplicate skill. No API key.