VibeFuse Documentation — Free AI Agent Harness for Windows

What's new in v0.1.23

• Local Servers tab — fourteenth Settings tab for Piper endpoint / voice / speed / volume, Whisper endpoint + model picker, HyperFrames Studio start/stop, Skill Seekers Connect, and Crawl4AI + SearXNG endpoints.
• HyperFrames Studio — HTML-to-video composer on the canvas at http://127.0.0.1:4567. Workspace C:\vibefuse\hyperframes. Vibe Agent can compose, check, preview, and render MP4s, then spawn a Video widget.
• Skill Seekers — local Python MCP cloned to C:\vibefuse\Skill_Seekers. Search docs-to-skill presets (React, Godot, Blender, Claude Code, Unity, …) before writing a new skill. No API key.
• Paid marketplace — price widgets and skills ($0.50–$500). Connect Stripe in Account → Stripe Connect; sellers keep 80%, Fuse Intelligence takes a 20% platform fee.
• GitHub, Copilot, Discord, Linear MCP — branded Tools cards alongside Google Workspace / Drive, plus HyperFrames, Skill Seekers, and Crawl4AI + SearXNG.
• Quieter voice replies — Piper speaks a short start cue when tools begin, then one spoken final answer. No TTS between tool calls.
• Linear widget — issue and project queries stay under Linear’s GraphQL complexity cap so the widget loads instead of failing.
• Browser tabs — click tabs to switch pages even when the canvas is zoomed or panned. The agent can jump to an already-open URL instead of fighting the live page.
• GitHub, Discord, and Linear MCP — list GitHub repos through the saved PAT, Discord default guild + snowflake IDs, plus Linear tools (from v0.1.10).
• VibeFuse Skills marketplace — browse and save skill packages on /marketplace. They sync into C:\vibefuse\skills and appear in the Skills rail. Publish from Account → Marketplace Settings.
• Vibe Agent creates and edits skills — ask the canvas chat to write a skill and it calls create_skill, update_skill, delete_skill, execute_skill, read_skill, and read_skill_file. The Skills section and Settings → Skills refresh immediately.
• Widget marketplace is live — browse VibeFuse widgets on the web, save to your library, and they sync into the desktop canvas automatically. Publish your own widgets from a local folder in Settings → Marketplace.
• Vibe Agent widget — a dedicated chat assistant on the canvas. It shares the voice assistant's tools: it can create, arrange, read, and remove widgets, type into terminals, drive browsers, web-search via local SearXNG, call MCP tools, create/edit skills, and speak short replies.
• Files widget — browse the session workspace with previews for images, video, audio, and PDFs; open or reveal in Explorer.
• Local voice engines — Whisper (whisper-cli.exe) and Piper (piper.exe) now run on-device under %LOCALAPPDATA%\VibeFuse\bin with the Jarvis voice; remote voice hosts become the fallback.
• Jarvis wake word — enable it in Settings → Voice and say “Jarvis”; VibeFuse plays a short “Yes, sir?” and takes your command hands-free.
• Voice notes mode — switch the dock to Notes, click to record, and VibeFuse summarizes the take into bullets and uploads it to your profile (Account → Notes).
• MCP Tools tab — two Google cards ship built in: Google Workspace (Gmail, Drive, Calendar, Docs, Sheets, Tasks, Slides, Meet, Contacts — ~24 tools) and Google Drive (read-only Drive MCP — ~3 tools). Same OAuth client, Connect / Sign in with Google / Use saved Google account, plus any custom stdio server.
• Provider grid — Settings → Agent now includes OpenAI, Ollama Local, Ollama Cloud, NVIDIA NIM, Featherless, Google Gemini, Groq, Together AI, OpenRouter, Plugsky, and Custom. Each card keeps its own key, endpoint, and model.
• In-app updates — Settings → Updates shows the release feed; Settings → About → Check for updates downloads and runs the installer update for you.
• Navigation editor — rename, reorder, and hide left-rail items; create your own categories (Settings → Navigation).
• Themes & polish — nine theme presets (Midnight Navy, Maroon & Gold, Obsidian, Blood Ember, Toxic Void, Arctic Blade, Copper Wraith, Neon Noir, Porcelain Light), per-session audio mute, voice history, and workspace state saved on close.

Overview — free widget canvas for the agent harness

VibeFuse (VibeFuse.exe) is Fuse Intelligence’s Windows AI agent harness for Windows 10+ x64. Agents, a system terminal, embedded browsers, a file explorer, and Vibe Agent chat each spawn as draggable widgets on a free-form React canvas. Voice is built in: hold-to-talk local Whisper STT, Jarvis wake word with Piper “Yes, sir?” ack, Notes mode, and near-real-time Jarvis TTS replies. Local Servers host Piper, Whisper, HyperFrames Studio, Skill Seekers, and Crawl4AI + SearXNG on your PC. The marketplace lets you sell widgets and SKILL.md packages through Stripe Connect.

• Free: every Fuse Intelligence account receives a profile-linked VF- key with no checkout
• Widgets: Terminal, Browser, YouTube, Files, Vibe Agent, Claude, Codex, Gemini, Cursor, Qwen, saved pages, and marketplace iframe widgets
• Skills: Claude/Codex-compatible SKILL.md packages in C:\vibefuse\skills — local folders, marketplace saves, and skills Vibe Agent writes for you
• Sessions: named canvases you can create, rename, clone, mute, and close without killing live PTYs
• Voice: hold-to-talk with local Whisper, Jarvis wake word, notes mode, and on-device Piper replies
• Tools: canvas-controlling assistant tools, skill CRUD tools, plus MCP servers (Google Workspace / Drive presets and custom stdio servers)
• Local Servers: Piper TTS, Whisper STT, HyperFrames Studio, Skill Seekers MCP, Crawl4AI scrape, and SearXNG search — all on-device
• Marketplace: live at /marketplace — browse, buy, save, sync, and sell widgets and Agent Skills (80% seller payout)

UI catalog — live desktop clones

Every showcase below is the same HTML/CSS as the VibeFuse desktop app: shell chrome, settings tabs, widgets, and overlays. Browse the full set (search + theme switcher) in the standalone component catalog.

Architecture

VibeFuse is a Windows desktop app. The main process owns PTY sessions, CLI and MCP management, local speech, marketplace sync, notes sync, in-app updates, IPC, and local persistence in C:\vibefuse\. The renderer is a React widget canvas with live terminals, embedded browsers, a file explorer, Vibe Agent chat, the Skills rail, the voice dock, and the thirteen-tab Settings modal. Licensing, version checks, marketplace, notes, and the update feed go through Fuse Intelligence; agent CLIs and voice engines run locally.

Requirements

Installer

VibeFuseInstaller.exe is a styled navy / steel / gold bootstrap (~90 KB). It reads the live version, downloads the desktop exe from /downloads/VibeFuse/, then installs anything missing before launch.

1. Open Downloads or Account → Downloads.
2. Copy your complimentary VF- key from Account → Product Keys.
3. Run VibeFuseInstaller.exe, paste the key, then approve the elevation prompt. Install will not start until the key verifies against your profile.
4. The installer reads version.json, then /api/app-version.php?product=vibefuse if needed.
5. It saves VibeFuse.exe under Program Files\VibeFuse (elevated) or %LOCALAPPDATA%\Programs\VibeFuse, and writes your verified license beside the exe.
6. 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.
7. The Whisper model (ggml-base.en-q5_1.bin) and the Jarvis Piper voice (jarvis.onnx) land in %LOCALAPPDATA%\VibeFuse\models\; the local speech engines (whisper-cli.exe, piper.exe) land in %LOCALAPPDATA%\VibeFuse\bin\.
8. Desktop and Start Menu shortcuts are created, then the app launches.

VibeFuse Setup×

VibeFuse v0.1.23 · Windows x64

Product key (required) VF-XXXX-XXXX-XXXX-XXXX Need a key? Get one free at fuseintelligence.org

1. Verify product key
2. Read latest version
3. Download VibeFuse.exe
4. Node.js runtime
5. Git
6. Claude Code CLI
7. Codex CLI
8. Gemini CLI
9. Cursor CLI
10. Qwen CLI
11. Whisper model
12. Jarvis voice + speech engines
13. Shortcuts
14. Launch VibeFuse

Installing Claude Code CLI…

Installing…

Updating from inside the app

VibeFuse checks /api/app-version.php?product=vibefuse (falling back to version.json). When a newer build exists, use Settings → About → Check for updates: the app downloads the installer to %TEMP%, runs it with --update --install-dir --license-key so the saved key is reused, closes itself, and the installer swaps the exe and relaunches. The Settings → Updates tab shows the same release feed as /updates/vibefuse.

Version

Current release in the database: v0.1.23. The same semver is stored in three places:

License key

VibeFuse is complimentary. On register, login, or Google sign-in the site creates an active free subscription and a unique VF-xxxx-xxxx-xxxx-xxxx key on your user id. 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 in the desktop app. It is saved beside VibeFuse.exe by the installer and kept in %APPDATA%\VibeFuse\license.json.

Workspace UI

The window is frameless with a theme-driven look. Drag the title bar to move it. The left rail is organized into three system sections: Agents (Vibe Agent first, then installed CLIs), Widgets (Browser, YouTube, Files, Marketplace action, saved pages, marketplace iframes), and Skills (every local and marketplace SKILL.md package). The canvas is a free-form plane.

Browser widgets remember URL and zoom in the widget config and share one browser session partition. Saved pages appear under Widgets without an address bar. Skills stay in their own section — hide one in Settings → Skills or Settings → Navigation without deleting the folder. Inactive session canvases stay mounted and hidden so terminals do not reset. Workspace state is flushed to disk when the window closes, so nothing is lost on quit.

Saving and loading canvas sessions

A session is a named canvas: widget layouts, sizes, z-order, zoom, pan, and per-session audio mute. Use sessions to keep a review layout, a coding layout, and a research layout without tearing down PTYs.

• + creates a new empty session
• Click a tab to switch. Hidden canvases keep their terminals alive
• × closes a session after confirm
• Tab menu (click … or right-click): Rename, Clone
• Clone copies layouts and config with new instance IDs (fresh PTYs)
• Each session can store its own workspace path in the left rail for terminal and agent cwd
• Each session has an audio mute toggle for its browser widgets
• Ctrl+Tab cycles sessions when the canvas is focused
• The workspace snapshot is persisted locally (C:\vibefuse\vibefuse.json, migrated from the old %APPDATA% state file) so sessions reload on next launch

Local speech to text, Jarvis wake word, and Piper speech

The title-bar voice dock is hold-to-talk. Pointer capture stays on the mic button so the take is not aborted mid-hold. Speech is local-first: recording goes to whisper-cli.exe (whisper.cpp, with ffmpeg for audio conversion) on your PC; if the local engine or model is missing, the take falls back to the remote Whisper host. Replies are spoken by piper.exe with the Jarvis voice (jarvis.onnx, British RP) on-device; the remote Piper host is the fallback. Default voice is jarvis at 1.45× speed.

While idle or thinking, only the mic button shows. During recording or Piper playback, a left-side equalizer animates beside the button. After transcription, the You / Assistant transcript panel appears to the right of the record button and fades after a few seconds (hover to keep it open). The dock has a mode picker below the orb — choose Normal Mode (hold-to-talk, with optional wake word) or Note-Taking (click-to-record). See Wake word and Note-Taking mode below.

The installer caches the Whisper model, the Jarvis voice, and both speech engines locally when missing; the app re-checks and repairs them on launch. Voice choices, speed, volume, the wake word toggle, and a Test voice probe live in Settings → Voice. Transcripts and replies are kept in Settings → History.

Jarvis wake word

The wake word makes VibeFuse hands-free. Turn it on once in Settings → Voice → Jarvis wake word and the app keeps listening in the background while it is in Normal Mode — no button, no hold.

1. Enable Jarvis wake word in Settings → Voice (Normal Mode only — the wake listener pauses in Note-Taking mode).
2. Say “Jarvis” — clearly, in a normal voice. Variants like “J.A.R.V.I.S”, “Jervis”, or “Jarviz” are recognized too.
3. VibeFuse plays a short cached “Yes, sir?” acknowledgement (Piper, no round-trip) and starts recording your command.
4. Speak your command, then pause. The take ends on silence and runs the same pipeline as hold-to-talk: transcription → provider reply → Piper speech — and canvas tool calls if the command asks for them.

Note-Taking mode

Note-Taking mode turns the voice dock into a one-click dictation recorder: take a note, and VibeFuse transcribes it, summarizes it into bullet points with an auto-generated title, and uploads it to your Fuse Intelligence profile.

Switching modes

Click the mode label under the mic orb (it reads Normal Mode or Note-Taking) and pick from the menu. You cannot switch mid-recording — finish the current take first. The menu summarizes each mode: Hold to speak · optional Jarvis wake word versus Click to record · summarize · upload (wake word paused).

Taking a note

1. Switch the dock to Note-Taking. Have a provider configured in Settings → Agent — the summary step uses it.
2. Click the orb to start recording (no holding). The equalizer animates while you talk.
3. Click again to stop. Very short takes are ignored with a hint to speak longer.
4. The dock shows Thinking while the take is transcribed locally, then your provider summarizes the transcript into bullets and a title.
5. The bullets and the full transcript upload to your profile, and the transcript panel confirms: Saved as “…” to your Fuse Intelligence notes.

What gets saved

• Each click-to-click recording becomes one note: bullet summary first, the full transcript below a divider
• Notes are grouped into sessions on your profile, so a run of notes reads as one stream
• If summarization fails (no provider, endpoint error), the raw transcript is still uploaded under a fallback title — nothing is lost
• Notes sync against your VF- key, so they follow your profile across machines
• Normal Mode never uploads — Note-Taking is the only path that writes to your profile, and the wake word is paused while it is active

Review everything at Account → Notes on the web, or in Settings → Notes inside the app, where you can browse, edit, and delete notes per session group.

Link your own AI provider

Set this in Settings → Agent. The live grid is eleven cards. Each card stores its own API key, endpoint, and model — switching a card restores the last values you used on that provider. Refresh models pulls the remote list when the API supports it; you can also paste a custom model id. Test connection pings the endpoint before you rely on it. The same provider powers Jarvis (voice) and the Vibe Agent chat widget.

The full Agent tab matches the desktop chrome — provider grid plus CLI install chips:

The voice assistant asks the model for JSON:

The dock shows the full text. Piper only speaks speech. The Vibe Agent widget skips the JSON envelope and replies in GitHub-flavored Markdown with tool calls (up to eight rounds per turn). If the endpoint is empty, the raw transcript is spoken through Piper.

MCP tools — Google, GitHub, Copilot, Discord, Linear

Settings → Tools is branded MCP cards plus your custom stdio servers. Google Workspace and Drive share one OAuth client. GitHub uses a PAT; Copilot uses a Copilot-capable token; Discord is a local Java MCP; Linear is hosted OAuth at mcp.linear.app. HyperFrames, Skill Seekers, and Crawl4AI + SearXNG also appear here and are started from Settings → Local Servers. The live pill shows ready · tool count. Jarvis and Vibe Agent call every connected tool the same way they call canvas tools.

• Enable a server and VibeFuse spawns it, lists its tools, and exposes them to the voice assistant and Vibe Agent as callable functions
• Servers stop cleanly with the app; the pill updates from connecting → ready / error
• Google sign-in is a sandboxed OAuth popup; the linked email prints on both cards
• If Drive/Gmail calls fail with “API has not been used…”, open Enable Workspace APIs for the shared OAuth project, then reconnect
• Built-in assistant search: web_search uses a local SearXNG endpoint (http://127.0.0.1:51452 by default)

The same Tools panel lives in Settings — Google Workspace, Drive, GitHub, Discord, Linear, and custom stdio:

Agent Skills — rail, Settings, and Vibe Agent

A skill is a folder with a Claude/Codex-compatible SKILL.md. VibeFuse loads every package from C:\vibefuse\skills into the left-rail Skills section and into Settings → Skills. Marketplace saves write the same folder. Vibe Agent can create, edit, delete, read, and execute skills without you touching disk.

What Vibe Agent can do

• Each Settings card shows name, Local or Marketplace badge, version, owner, description, Hide / Show, and Remove (confirm → Delete folder)
• Hide removes the skill from the rail without deleting the folder — same visibility toggle as Settings → Navigation
• Drop any folder with SKILL.md into C:\vibefuse\skills and it appears after a catalog refresh
• Connected provider agents (Claude, Codex, Cursor, …) can also read these files — they are standard Agent Skills, not VibeFuse-only

Publish or polish a skill from Settings → Marketplace or Account → Marketplace Settings. Step-by-step: Write a skill — the full guide.

Widget types

Built-in widgets come from the left rail. CLI widgets are live PTY sessions; Browser and YouTube are embedded Chromium panes; Files is a workspace explorer; Vibe Agent is a tool-using chat. Use the tabs to open each type.

Widget creation guide

A widget is a folder. VibeFuse scans the marketplace roots (and the app widgets/ tree) for manifest.json and serves the package over the widget:// protocol.

1. Create a folder, for example hello-card, inside the marketplace path from Settings → Widgets.
2. Add manifest.json with a unique id such as com.yourname.hello-card.
3. Add index.html (or set entry). The iframe runtime loads that file.
4. Restart VibeFuse or reopen Settings so the catalog rescans.
5. Spawn the widget from the rail / catalog once it appears.

Permissions you may declare: pty, shell, network, webview, fs.read, fs.write, clipboard, storage, voice. Only request what the widget needs.

Categories: cli, browser, utility, media, custom. Layouts use free-form x, y, w, h, z on the canvas. Keep WIDGET_PROTOCOL_VERSION compatibility in mind — the host API is versioned so marketplace packages stay loadable.

Ready to share it? Publish the folder straight from Settings → Marketplace (see below). Uploads accept text files (html, css, js, mjs, json, svg, md, txt, map) up to 750 KB each, and your profile name is stamped as the author.

Marketplace — widgets & Agent Skills Live

The marketplace is live at /marketplace — discover widgets and Agent Skills on the web, save them to your library, run them on the canvas, and publish your own from the desktop app or polish them in Account → Marketplace Settings.

On the web — /marketplace

• Live pulse — trending widgets, trending skills, trending sellers, and staff picks in a rotating spotlight (click the dots in the clone above)
• Widget catalog — every published widget in one grid; each card shows category, saves, and Free tag
• Skills catalog — #skills on the same page; each skill is a savable SKILL.md package with a detail page at /marketplace/skill/{slug}
• Widget pages — banner, screenshot slider with captions, rich “About this widget” description, version/category/saves/views meta, and Save to VibeFuse
• Skill pages — description, SKILL.md preview, owner, version, and Save to VibeFuse (writes the skill into your library)
• Sellers — seller cards with follower and listing counts; follow creators to catch their next widget or skill drop
• Your library — the hero shows how many widgets and skills you have saved while signed in

Saving a widget downloads it into C:\vibefuse\widgets and lists it under Widgets. Saving a skill downloads SKILL.md + manifest.json into C:\vibefuse\skills and lists it under Skills. Sync runs at launch and when you close the in-app marketplace webview.

In the app — browse & sync

• The Marketplace action under Widgets (or Settings → Marketplace → Open marketplace) opens the browse modal — the full website in a themed webview
• Closing the modal triggers an immediate library sync of widgets and skills; the app also syncs at launch
• Settings → Marketplace shows your linked license, widgets folder, skills folder, local packages (including saved skills), and live sync results

Publish a widget from Settings → Marketplace

1. Link a valid VF- key (the tab shows your seller profile; there is a shortcut to Product Keys if not).
2. Build the package: a folder with manifest.json + entry file (see the creation guide).
3. Open Settings → Marketplace. Local widgets scans your widgets folder — each card shows its status: Not on marketplace, Uploaded · unpublished, or Live on marketplace.
4. Click List on marketplace to publish a new widget, or Update listing to push new files and metadata for one already listed. Choose other folder publishes any folder outside the marketplace directory; Reveal folder opens it in Explorer.
5. Uploaded-but-unpublished listings show a Publish all (N) banner — one click makes every hidden listing live.
6. Watch results in the tab: your listings show version, status, saves and installs counts, with Unpublish / Publish buttons per widget and a View link to the public page.

Uploads accept text files (html, css, js, mjs, json, svg, md, txt, map) up to 750 KB each; your profile name is stamped as the author automatically. Republishing the same folder updates the listing in place.

Manage your listings on the website — Account → Marketplace

The seller dashboard is where listings get polished after upload. It lists every widget and skill on your profile (published and unpublished):

• Widgets — name & descriptions, card image & banner, screenshot slider with captions, Published / Unpublished, saves / views
• Skills — create a new skill, edit slug / short description / SKILL.md body (paste markdown or upload a .md file), Save draft, Publish, Unpublish
• Stats — saves, detail views, and a link to the live public page for each listing

Local Servers — Piper, Whisper, HyperFrames, Skill Seekers, crawl

Settings → Local Servers is the fourteenth Settings tab. It is where on-device engines live: Piper TTS, Whisper STT, HyperFrames Studio, Skill Seekers MCP, and the Crawl4AI + SearXNG web stack. Voice wake-word and mic/speaker devices stay on Settings → Voice; this tab owns endpoints, models, and start/stop.

Piper TTS

Whisper STT

Models live under %LOCALAPPDATA%\VibeFuse\models\ as ggml-*.bin. Tiny English is the live-voice default; the installer also caches ggml-base.en-q5_1.bin. ffmpeg on PATH is required for microphone conversion.

HyperFrames card

• Workspace C:\vibefuse\hyperframes · Studio port 4567
• Buttons: Start Studio / Restart / Stop / Open widget
• State pill: ready / connecting / error / disabled. Needs Node.js 20+.
• Chips: Studio UI, Init, Preview, Render, Capture

Skill Seekers card

• Local Python MCP from yusufkaraaslan/Skill_Seekers at C:\vibefuse\Skill_Seekers
• Connect / Disconnect — VibeFuse clones the repo, creates a venv, and starts FastMCP. Requires Python 3.11+ (3.12 preferred). No API key.
• Chips: Find skills, Docs → skill, GitHub repo → skill
• Starts with the desktop app so Vibe Agent can search presets before calling create_skill

Crawl4AI + SearXNG

• SearXNG endpoint — default http://127.0.0.1:51452 for web_search / firecrawl_search
• Crawl4AI endpoint — default http://127.0.0.1:11235 for scrape / map / crawl (local Python sidecar, no Docker, no Firecrawl Cloud)
• Start web tools / Refresh — starts the sidecar; if it is down, scrape still uses a built-in fetch fallback

HyperFrames — HTML-to-video on the canvas

HyperFrames is a local HTML-to-video studio. VibeFuse starts it with npx hyperframes@latest, binds the workspace root so the Renders tab lists C:\vibefuse\hyperframes\renders, and embeds Studio in a widget. Ask Vibe Agent or Jarvis to make a video about the current workspace folder — it inspects README / package.json, composes scenes, checks, renders an MP4, and can spawn a Video widget.

Bundled skills such as hyperframes-core, hyperframes-animation, hyperframes-creative, hyperframes-cli, and media-use live in the Skills rail so Claude / Codex / Cursor widgets can follow the same composition contract. See the HyperFrames landing page.

Skill Seekers — find a skill before you invent one

Skill Seekers is a local Python MCP that scrapes documentation and packages Claude-compatible SKILL.md folders. VibeFuse clones the repo to C:\vibefuse\Skill_Seekers, builds a venv, and exposes tools to Vibe Agent. The desktop lookup skill (skill-seekers-lookup) tells the agent to call list_configs first so you do not duplicate React, Godot, Blender, Claude Code, Unity, or other presets.

1. Install Python 3.11+ (3.12 recommended) and Git. The app looks in %LOCALAPPDATA%\Programs\Python\Python312 and C:\Python312.
2. Open Settings → Local Servers → Skill Seekers → Connect. First run clones and installs; later launches reuse the venv.
3. Ask Vibe Agent: “Is there already a skill for React / Godot / this docs site?” It runs skill-seekers-lookup then Skill Seekers MCP tools.
4. To build a new skill from docs: generate a config, scrape, then package. Packaged output can land under C:\vibefuse\skills.
5. If tools are missing, reconnect the card. Python must use UTF-8 stdio (VibeFuse sets PYTHONUTF8=1 so emoji logs do not crash on Windows cp1252).

No cloud API key. The MCP HTTP port is 8765 on localhost. Publish a finished skill from Account → Marketplace Settings or Settings → Marketplace — free or paid. See Skill Seekers and Sell AI skills.

Sell widgets & skills — Stripe Connect, 80% payout

The marketplace is a creator economy, not just a free dump. List a widget or Agent Skill as Free or paid ($0.50–$500.00). Buyers check out through Stripe. You receive 80% of each sale; Fuse Intelligence keeps a 20% platform fee. Payouts go to your connected Stripe account.

1. Sign in and finish a public seller profile (display name, avatar, optional bio / storefront banner) in Account → Marketplace.
2. Open Account → Stripe Connect and complete onboarding until charges and payouts are enabled.
3. Publish a widget from Settings → Marketplace (folder with manifest.json + entry) or a skill from Account → Marketplace Settings (slug, short description, SKILL.md body).
4. Set a USD price. Paid prices require Stripe to be fully enabled. Leave $0.00 for Free.
5. Polish the listing: card image, banner, screenshot slider + captions, long description. Publish.
6. When someone buys, they get the package in their VibeFuse library; you see net earnings on Account Overview (7-day seller chart) and in Stripe.

• Minimum paid price $0.50 · maximum $500.00 · two decimal places
• Application fee is taken at checkout — you never invoice buyers yourself
• Unpublish hides the listing without deleting the package folder
• Followers see your next drop on the marketplace pulse and your storefront

Account, profile settings, and the community forum

The same Fuse Intelligence account licenses VibeFuse, syncs notes, owns your marketplace library, and posts on the forum. Nothing here is a second login.

Account Overview & settings

• Display name, theme, accent — site chrome (dark/light + hex accent). Desktop themes stay in Settings → Appearance (nine presets).
• Avatar — used on the forum, seller cards, and marketplace storefront. Clear or upload from Account settings.
• Forum signature — appended to your posts.
• Notes email + cloud notes — where VocalFuse / VibeFuse Note-Taking summaries can be delivered; cloud notes toggle for Account → Notes sync.
• Product Keys — complimentary VF- key. Required for desktop marketplace sync, publish, and notes upload.
• Downloads — installer and exe for the linked license.
• Stripe Connect — seller onboarding for paid listings.
• Marketplace / Skills marketplace — polish listings, set prices, publish, unpublish, view saves / views / earnings.
• Overview chart — 7-day marketplace views or net seller earnings after the platform fee.

Forum

• Hub: /community · threads: /forum
• Public forums (not staff-only), thread pages, user profiles at /forum/u/{id}, ranks at /forum/ranks (Spark through Apex, plus Administrator crown)
• Notifications (bell) for replies, marketplace saves/purchases, and staff messages — /api/notifications
• Discord: same community, live chat — linked from the hub

Seller storefronts and forum profiles are public. Account settings pages are noindex — they are not SEO targets.

Canvas control — let Jarvis design your layout

Jarvis (the voice assistant, by wake word or hold-to-talk) and the Vibe Agent chat widget can both redesign the canvas on command. Ask in plain language and the assistant picks the right tool — spawning widgets, tiling them across the view, resizing, renaming, stacking, and clearing, all on the active session.

What the assistant can do

Good to know:

• Layout changes apply to the active session and respect the current zoom/pan view — “fill the view” means the visible canvas, not the whole virtual plane
• Spawning up to 6 CLI/terminal widgets or 4 browser widgets per call; layouts can be passed as layouts[] or a widgets[] batch
• The assistant always calls list_canvas_widgets first when it needs to find widgets by name
• Prefer the no-code side? Settings → Navigation edits the rail manually, and Settings → Appearance redesigns every surface with nine theme presets
• Richer scripted theming surfaces (per-widget styling APIs) remain planned

Settings

Open the gear in the title bar. Fourteen tabs, grouped Workspace / Assistants / Library / System. The sidebar has a Search field that filters tab names.

Troubleshooting

Related