promptapp.run

POC/Experimental

A static single-page app that compresses an entire AI micro-app into a shareable URL. The full definition of the app -- its name, personality, system prompt, accent color, configuration -- gets LZ-compressed and packed into the URL hash fragment. No server, no database, no deploy. The URL is the app. This repo is a proof of concept for that idea, with 30 sample apps you can open and use right now. It runs on GitHub Pages.

The idea

Every useful AI interaction starts as a prompt. But the gap between "I wrote a good prompt" and "other people can use my prompt" normally requires deploying a backend, managing auth, standing up infrastructure. promptapp.run collapses that gap to zero.

A system prompt plus some metadata (name, tagline, accent color, input placeholder) compresses to roughly 500-2000 characters of URL. That's well within browser limits. The compressed payload lives in the hash fragment, which browsers never send in HTTP requests, so the app definition travels entirely client-side. No server ever sees it.

This also means full transparency. Anyone with the URL can inspect the system prompt that powers it. There's no black box. If you share one of these apps with someone, they can read every word of the instructions driving the AI.

Try it

Open the live site: dvelton.github.io/promptapp-runner

All 30 built-in apps work immediately with simulated responses. No API key, no setup, no account required. Just pick an app and start using it.

How URL compression works

The app payload is a JSON object:

{
  "name": "Contract Decoder",
  "tagline": "Plain English for legal fine print",
  "accent": "#2563eb",
  "inputLabel": "Paste a contract clause",
  "inputPlaceholder": "e.g. an indemnification provision",
  "systemPrompt": "You are a legal plain-language translator..."
}

That object gets:

1. Serialized to a JSON string
2. Compressed with LZ-String
3. Encoded as URI-safe base64
4. Appended to the URL hash: #app=<compressed>

The hash fragment never leaves the browser. Decompression happens entirely client-side.

Sample apps

The gallery includes 30 apps across six categories:

• Legal and Contracts -- Contract Decoder, NDA Reviewer, and more
• Career and Negotiation -- Offer Negotiator, Resume Bullet Rewriter, and more
• Writing and Communication -- Email Vibe Check, Tone Adjuster, and more
• Developer Tools -- Regex Explainer, Error Message Decoder, and more
• Business and Strategy -- Pitch Deck Feedback, OKR Writer, and more
• Personal and Fun -- Explain Like I'm 5, Recipe from Leftovers, and more

Each app has its own system prompt, accent color, and themed interface. Each also has pre-baked sample responses so you can see what the interaction looks like.

Create your own

Click "Create your own" on the home page. Fill in a name, tagline, accent color, and system prompt. Click "Generate Share Link" and you get a URL you can send to anyone. The recipient opens the link and gets a fully themed chat interface built around your prompt -- no installation, no sign-up, nothing to configure on their end.

Self-host

Fork this repo, go to Settings, then Pages, select GitHub Actions as the source, and you're done. Your instance is live at https://<username>.github.io/promptapp-runner/.

Or copy index.html to any static host -- S3, Netlify, Vercel, Cloudflare Pages, a bare nginx server. One file, no build step, no dependencies beyond two CDN links loaded at runtime.

Transparency

System prompts are embedded in share URLs. Anyone can click Inspect to see the full prompt powering any app. This is by design. If you're sharing an AI tool with someone, the people using it should be able to see what it does.