The simple way to turn an idea into a working app or website.
Describe what you want in plain English, watch it get built in seconds, and use it right away.
Documentation · Download · Discord · License
AppBlips.Demo.mp4
AppBlips is built for everyone. Whether you've never written a line of code or you ship production apps every day, AppBlips turns plain-English prompts into working apps, games, and websites in seconds.
It runs locally on your machine—bring your own API key (Anthropic, OpenAI, Gemini, DeepSeek, OpenRouter, or local LM Studio/Ollama) or use built-in AppBlips AI Credits. No mandatory accounts, no cloud lock-in, and the generated code is always yours to inspect, export, or host anywhere.
Every studio shares the same fast loop: Prompt → Live Preview → Version History → Export/Publish.
| Studio | Best For | Under the Hood |
|---|---|---|
| App Studio | Interactive tools, utilities, and dashboards | JavaScript-driven, persists data in the browser, defaults to mobile preview. |
| Website Studio | Landing pages, portfolios, business sites, and blogs | Full website anatomy with Swiper, GSAP, or Leaflet. Supports both AI edits and direct click-to-edit. |
| Game Studio | Playable 2D browser games with touch & keyboard controls | Raw Canvas 2D for simple arcade games or Phaser 3 (via CDN) for physics, sprites, and tilemaps. |
- Instant Live Preview — Watch your app, game, or website render right next to your prompt in a safely sandboxed view, with mobile/desktop toggles and zoom controls.
- Click-to-Edit & Chat Refinements — Click text, images, and links directly in Website Studio to make quick changes, or ask the AI in plain English for deeper structural tweaks.
- Project Image Gallery ("My Images") — Upload your own photos and logos to a project's gallery and tell the AI how to use them (e.g., "put team-photo in the hero"). Reference screenshots are also supported.
- Local Voice Input — Speak your prompts using the built-in microphone. Powered by a bundled Whisper model running locally on your hardware—zero audio is uploaded.
- Automatic Versioning — Every generation and edit is saved as a snapshot you can undo/redo at any time.
- Bring Your Own Key or Use Credits — Connect your own API keys (or local models) for free, or subscribe to AppBlips AI Credits (starts with a 3-day, 100-credit trial; converts to the $10/mo Starter Plan for 1,000 credits/mo unless cancelled).
- One-Click Connectors — Wire up Web3Forms (contact emails), Umami/Plausible (analytics), tawk.to (live chat), Square (payments), or Pollinations (AI image generation & chat) via Settings → Connectors.
- Single-File HTML Export — Download any generated project as a self-contained HTML file ready to share or host anywhere.
- One-Click Web Publishing — Push your project live to a free
*.surge.shHTTPS address (with optional custom favicon) directly from the preview toolbar. - Built-in Showcase — Browse community websites and games directly inside the app (even offline), and submit your own published creations.
🎉 What's New in v0.1.15 (Click to expand)
- AppBlips AI Credits: Build without an API key. Start a 3-day free trial (100 credits; requires a valid payment method and converts to the $10/month Starter Plan for 1,000 credits/mo unless cancelled in Settings → Account & Usage) or pick a monthly plan (Starter, Builder, or Studio). Your own saved API provider always takes priority if configured.
- My Images Gallery: Upload project photos and logos, ask the AI to place them by name, and have them automatically bundled on export or publish.
- Showcase & Site Icons: Browse real AppBlips projects from the studio picker (even offline) and attach custom favicons when publishing to Surge.
- Quality of Life: Streamlined first-run setup, a refreshed Light theme with higher contrast, Full-size preview support for websites, external link confirmation prompts, a manual Check now update button in Settings → Workspace, and Electron 44.9.1 security updates.
For older releases, see the full changelog or GitHub Releases.
- Windows: Install from the Microsoft Store (signed by Microsoft with automatic updates).
- Linux: Download the latest
.AppImage(any distro) or.deb(Debian/Ubuntu) from GitHub Releases. (For AppImage: right-click → Properties → allow executing as a program). - macOS: No native installer yet—use Docker or run from source below.
First run: Choose where to save your projects, then either connect AppBlips AI Credits or paste an API key (Anthropic, OpenAI, Gemini, DeepSeek, OpenRouter, or local LM Studio/Ollama). Keys are encrypted using your OS system keychain.
With Docker Compose v2 installed, run from the project root:
docker compose up -dOpen http://localhost:3000. Stop with docker compose down, or update with git pull && docker compose up -d --build.
Requires Node.js 22.12 LTS or newer.
git clone [https://github.com/techmitten/app-blips.git](https://github.com/techmitten/app-blips.git)
cd app-blips
npm install
npm startOpen http://localhost:3000.
- Want the Electron desktop app? Run
npm run desktop(dev) ornpm run desktop:distto build installers intodist-desktop/. - Optional
.envsetup: Copy.env.exampleto.envto pre-configure providers or server settings. See the Configuration Reference.
Full guides and references live at docs.appblips.com:
- Introduction · Desktop Quickstart · Run from Source · Docker Guide
- Studio Feature Guides · Publishing to Surge · Community Showcase
src/
App.jsx # Workspace/generation state, undo-redo, composition root
previewBridge.js # Script injected into generated apps to bridge the sandboxed iframe
lib/ # Framework-free logic: LLM calls, surgical edits, prompts...
hooks/ # Stateful concerns: projects, preview viewport, theme...
components/ # Presentational UI: Header, BuildPanel, PreviewPane, modals...
electron/ # Desktop app: main process, preload, on-disk project store,
# encrypted provider settings, and the in-process /api/chat handler
server.js # Browser version (npm start, Docker): serves dist/ and /api/chat
Dockerfile # Container image for the browser version (docker-compose.yml)
testing/ # Standalone Node scripts for exercising the proxy and preview
scripts/ # Build helpers, including fetch-whisper-model.js for voice input
🔧 Deep Dive: Subsystems & Testing (Click to expand)
For a complete architectural walkthrough (generation flow, preview sandboxing, and LLM proxy internals), see CLAUDE.md.
- Model & Runtime: Uses an 8-bit quantized ONNX conversion of OpenAI's multilingual
whisper-base(~76 MB) via@huggingface/transformers(WASM backend) inside a dedicated Web Worker (src/lib/whisperWorker.js). - Offline Bundling:
scripts/fetch-whisper-model.jspulls pinned model files intopublic/whisper/during build/dev and verifies SHA-256 hashes. No external network calls occur at runtime. - Permissions:
electron/main.jsrestrictsmediaaudio permissions strictly to the top-level app UI—never the sandboxed preview iframe, and never the camera. - Test:
node testing/testVoiceInput.mjs(afternpm run build).
- Flow: Compiles JSX, injects loop protection, strips the preview bridge, bundles referenced Pollinations/gallery images under
images/plusfavicon.png(src/lib/favicon.js), and uploads a gzipped tarball to Surge viaelectron/server/surgeApi.js. - Auth Storage: Passwords are never saved. The returned Surge token is encrypted in the system keychain on desktop (
electron/surgeStore.js) or saved with0600permissions insurge-token.jsonfor browser/Docker builds. - Test:
node testing/testSurgeApi.js(offline; Surge is stubbed).
- Storage & References: Image bytes live in a content-addressed blob store (
blobs/on desktop, IndexedDB in browser), while project metadata tracks[{ hash, name, type, width, height, size, note }]. - Privacy & Export: The AI only receives image file paths, dimensions, and user notes (
buildImageGalleryNoteinsrc/lib/prompts.js)—never raw pixels. At render time,images/<name>swaps toappblips://assetsordata:URLs, and bundles cleanly on export/publish. - Test:
node testing/testGalleryImages.js.
- Billing & Auth: Managed via Polar (
electron/server/creditsApi.js,src/lib/credits.js). Users authenticate via emailed one-time codes (/api/credits/connectand/api/credits/verify). Credits route requests to DeepSeek (~4,000 weighted tokens per credit; output weighted 3x). - Host Relay vs. Direct: Packaged apps relay through
appblips.com; self-hosters can setPOLAR_TOKEN(withorders:read) andDEEPSEEK_API_KEYin.envfor direct mode. Tokens are stored encrypted on desktop (credits.json) or0600(credits-key.json) on server. - Test:
npm run test:credits(addPOLAR_TOKENandAPPBLIPS_POLAR_CONTRACT=1for live contract verification).
- Integration: Uses Pollinations' device flow (
electron/server/pollinationsApi.js). Keylessgen.pollinations.aiURLs are resolved server-side after build into publicmedia.pollinations.aiassets and bundled on export. - In-App AI: Injects
window.Pollinations(chat,image,mountConnect) viasrc/lib/pollinationsHelper.js. Supports visitor-paid device flow (default) or a budget-capped site key. - Test:
node testing/testPollinations.js(offline; stubbed).
- Strict Sandboxing: Generated apps run inside a sandboxed iframe (
sandboxwithoutallow-same-origin) using an opaque origin, isolating them completely from the parent app. - Local-First Keys: The desktop app opens zero network ports and encrypts API keys in your OS keychain. The browser/Docker server binds strictly to
127.0.0.1(localhost) and stores keys on disk (~/.appblipsorappblips-datavolume with restricted permissions)—never in browser storage. - Opt-In Cloud Services: Nothing leaves your machine unless you connect an external provider, opt into AppBlips AI Credits (which routes prompts through AppBlips' proxy to DeepSeek), or explicitly click Publish to upload static files to Surge.
Licensed under the MIT License. Third-party open-source licenses are available inside the app under Settings → Licenses.