The AI workspace where every model, agent, voice, and task lives under one roof.
Chat with every model, convene agent councils, code in terminal swarms, talk hands-free, schedule AI actions, and keep everything in one local-first memory.
VibeSpace is the desktop app. Jarvis is the built-in assistant inside it — voice, command bar, actions, and calling. Jarvis is not the product name.
⬇ Download latest · 📦 Install guide · 🖼 Screenshots · 🛠 Developer setup · 🌐 Website · ▶ Live demo
| Overview | Screenshots | Features | Dictation | Install | Security | Dev setup | Testing | Roadmap |
|---|
flowchart TB
VS[VibeSpace desktop]
VS --> T[Terminal swarm]
VS --> J[Jarvis voice + assistant]
VS --> D[Ctrl+Space dictation]
VS --> S[Schedule Jarvis Actions]
VS --> A[Agent council]
VS --> M[21+ model providers]
VS --> MEM[Persistent local-first memory]
T --> OC[OpenCode · Claude Code · Codex · shell]
J --> K[Kokoro local · Deepgram cloud]
| 🖥️ Terminals Live PTY grid · agent CLIs one-approval orchestration |
🎙️ Jarvis voice Local Kokoro · hands-free stop-mid-reply orb |
⌨️ Ctrl+Space Dictation in-app and into any desktop app |
🗓️ Jarvis Actions Scheduled prompts that actually run & save output |
Every image below is real app media. Workspace shots are captured from the actual app bundle with
scripts/capture-screenshots.mjs; the onboarding shot is from the real Linux desktop (Tauri) build. Desktop-only surfaces are captured on real machines — workflow indocs/MEDIA_CAPTURE.md.
Ctrl+Space dictation — VibeSpace STT, never Win+H |
Desktop build — real Tauri app, first launch |
Schedule & Jarvis Actions — outputs open in-page |
Context Map — project folders become AI context |
Terminal grid — live PTYs + agent CLIs |
Jarvis voice — personas, Kokoro local, Deepgram cloud |
More screenshots — voice engines, plans
No videos are committed yet — we don't fake media. The exact capture workflow (targets:
Ctrl+Space dictation in-app and into Notepad/VS Code, the Jarvis stop-response orb, a
scheduled Jarvis Action firing, the 10-terminal orchestration approval card, Context Map
node inspection, and the website "Call Jarvis" demo) lives in
docs/MEDIA_CAPTURE.md.
Status legend: ✅ Ready (code-complete, covered by the automated suite) · 🖥️ Desktop validation needed (works in code/tests; needs real mic/OS pass) · 🔭 Planned
| Feature | What it does | Status | Notes |
|---|---|---|---|
| Chat + Composer | Streaming chat with every wired provider, model picker, attachments, slash commands, queued messages | ✅ Ready | Mock provider works with zero keys |
| Jarvis Voice | Hands-free or push-to-talk, personas, stop-mid-reply orb, visible paused state | 🖥️ Desktop validation | Needs a real microphone pass |
| Ctrl+Space Dictation | Focus-aware: dictates into the focused in-app input, or a small overlay pastes into external apps | 🖥️ Desktop validation | Shared chat STT pipeline · never Win+H |
| Kokoro local voice | ~89 MB neural TTS, downloads once on first use, runs on-device, free on every plan | 🖥️ Desktop validation | Checksum-verified, resumable download |
| Hive pipeline | Multi-model sequential refinement (Hive Balanced), chat-only | ✅ Ready | Hosted path needs a paid plan; BYOK works |
| Council mode | Side-by-side agent panels with broadcast + synthesize | ✅ Ready | Requires 2+ agents (guided if missing) |
| Agents / Subagents | Create agents with a step-by-step wizard; /multitask + /subagents panels with real statuses |
✅ Ready | Panel groups agents vs subagents |
| Skills | Reusable instruction bundles, /skills picker, markdown editor with safe preview |
✅ Ready | |
| Schedule Jarvis Actions | Scheduled prompts that execute on their saved model; outputs open inside Schedule; recurrence + dedupe | ✅ Ready | Runs while the app is open |
| Terminals | 10-pane PTY grid, persistence, one-approval multi-terminal orchestration with AGENTS.md briefings | 🖥️ Desktop validation | PTYs need the desktop app |
| Context Map | Interactive AI map of a project folder; attaches to chat, informs terminal agents | ✅ Ready | Full file access needs the desktop app |
| Ollama / local models | API-first connect (no tray needed), model list, pull progress, offline routing | 🖥️ Desktop validation | Daemon paths verified live; inference needs real hardware |
| Plugins | GitHub, Figma, Supabase, Shopify, Slack + mock implemented; real connection tests | ✅ Ready | Catalog extras are generic/configurable — no fake connected states |
| Settings | 15+ sections, persisted, honest unavailable states, OS keychain for keys on desktop | ✅ Ready | |
| Billing / Plans | Stripe checkout + customer portal via authenticated edge functions; server-enforced budgets | 🖥️ External verification | App-side flow inspected; dashboard/server verification pending |
| AI calling (Jarvis Call) | In-app WebRTC + real phone calls via Twilio cloud | 🔭 Server hardening | Needs phone-jarvis plan enforcement deploy |
| Security / Privacy | Local-first storage, key redaction, approval-gated actions | ✅ Ready | See Security & privacy |
One hotkey, focus-aware, fully VibeSpace-owned:
| Where you are | What happens |
|---|---|
| Inside VibeSpace (composer, agent prompts, settings fields) | Dictation goes straight into the focused input — no floating overlay over the app |
| Outside VibeSpace (browser, VS Code, Notepad, games) | A small VibeSpace overlay listens, transcribes, and pastes into the focused app where OS input permissions allow |
- Shared STT pipeline — both paths use the same engines as VibeSpace chat: local faster-whisper → built-in speech recognition → Deepgram → Groq (order follows Settings → Speech to Text)
- Never Win+H — VibeSpace does not route dictation through Windows default dictation
- Real states — listening, transcribing, pasting, error — with Retry/Clear and clear fix paths for missing keys, mic permission, or paste failures
- Composer mic —
Ctrl+CapsLockor the composer mic button dictates into the chat input with live partial transcripts
Note
Release status: the latest published installers are v0.1.45; source is at v0.1.48. New signed Windows/macOS binaries are pending — the one-line installers always fetch the newest published release. Windows builds ship with updater signatures; Authenticode/notarization for public distribution is still in progress (see Roadmap).
| Windows 10/11 | macOS 12+ / Linux |
|---|---|
irm https://raw.githubusercontent.com/Cookie774-GameDev/VibeSpace/main/install/install.ps1 | iexSilent per-user NSIS install + |
curl -fsSL https://raw.githubusercontent.com/Cookie774-GameDev/VibeSpace/main/install/install.sh | bashmacOS: DMG → |
Or grab installers directly from the latest release
— .exe, .msi, .dmg, .deb, .rpm, AppImage — and verify against SHA256SUMS.txt
(details + troubleshooting in DOWNLOAD.md).
Honest, current state of the codebase and validation:
| Area | Status |
|---|---|
| Typecheck · tests · builds · installers · secret scans | ✅ Passing (see Testing status) |
| Linux install + desktop boot | ✅ Verified — AppImage install, plus the real Tauri build booted and ran in a Linux VM |
| Windows install + GUI | 🖥️ Hardware validation needed (scripts parse-validated; asset URLs verified) |
| macOS install + GUI | 🖥️ Hardware validation needed (install script dry-run + DMG SHA-256 verified) |
| Real-microphone voice & dictation | 🖥️ Hardware validation needed |
| Ollama inference | 🖥️ Real desktop needed (daemon API verified live; inference crashed only in the CI sandbox) |
| Code signing / notarization | 🔭 Pending before public binaries |
| Supabase / Stripe server-side | 🔭 App-side flow inspected in code; external dashboard verification pending |
- Local-first — chats, tasks, schedules, terminals, and settings live in IndexedDB/Dexie on your machine; cloud sync is opt-in
- API keys — stored in the OS keychain on desktop (session-only memory in browser preview); never logged, never sent to terminals, redacted in the dev console
- Dictation — no Win+H dependency; audio goes only to the STT engine your settings selected (local engines keep it on-device); no raw microphone audio is stored
- Approval-gated actions — destructive Jarvis actions (terminal orchestration, file writes, schedules) show approval cards; decline does nothing; malformed plans fail closed
- Terminal safety — agent briefings are delivered through
AGENTS.mdfiles, never typed into your shell; scrollback is sanitized and bounded - Payments — no Stripe secrets in the frontend; checkout/portal run through authenticated Supabase Edge Functions; webhooks verify signatures with idempotency
- Markdown safety — skill previews escape all HTML and only allow safe image schemes
Prerequisites: Node 20+, Rust 1.85+, and the Tauri prerequisites for your OS.
git clone https://github.com/Cookie774-GameDev/VibeSpace.git
cd VibeSpace
npm install
npm run tauri:dev # desktop app (web-only preview: npm run jarvis)Linux desktop dependencies
sudo apt-get install -y libwebkit2gtk-4.1-dev libgtk-3-dev \
libayatana-appindicator3-dev librsvg2-dev libasound2-dev \
libstdc++-13-dev g++ # last two needed for the kokoro voice featureAll checks
npm run typecheck # TypeScript
npm --prefix app run test # Vitest suite
npm run build # production web build
cd app/src-tauri && cargo check && cargo test --lib # Rust
npm run test:release-manifest # updater manifest
bash -n install/install.sh # installer syntax
node scripts/boot-validation.mjs # headless boot check (serve app/dist first)
node scripts/capture-screenshots.mjs # refresh README screenshotsSee SETUP.md for full prerequisites and CHANGELOG.md for history.
Verified on this branch (v0.1.48):
| Check | Result |
|---|---|
| TypeScript typecheck | ✅ clean |
| Vitest suite | ✅ 880 tests / 165 files passing |
| Production web build | ✅ passing |
cargo check + cargo test --lib |
✅ passing (10/10) |
| Release-manifest test | ✅ passing |
| Installer scripts (bash + PowerShell parse) | ✅ clean |
Website JS/CSS + asset checks (incl. /billing pages) |
✅ all 200 |
| Secret scan (repo + diff) | ✅ clean |
| Headless boot validation (routes + settings, console errors) | ✅ passing |
| Real-hardware validation (mic, Windows/macOS GUI) | 🖥️ outstanding — see Launch status |
Planned or in progress — not shipped yet:
- Signed public binaries — Windows Authenticode + macOS notarization before wide release
- Real hardware validation pass — microphone voice/dictation, Windows/macOS GUI installs, game/fullscreen dictation
- Phone-Jarvis server-side plan enforcement — call endpoints currently verify identity but not tier/budget; deploy pending
- Hosted per-tier model allowlist and hosted single-model chat metering (
message-complete) - External Supabase/Stripe dashboard verification
- Hive Custom preset editor — presets beyond Balanced are internal
- VibeBench, mobile companion, team canvases — vision stage
| Website | vibespaceos.com |
| Live demo (GitHub Pages) | cookie774-gamedev.github.io/VibeSpace |
| Releases | github.com/Cookie774-GameDev/VibeSpace/releases |
| Issues | github.com/Cookie774-GameDev/VibeSpace/issues |
| Media capture workflow | docs/MEDIA_CAPTURE.md |
VibeSpace — built for vibe coders, by a vibe coder.
Apache 2.0 · License