# RēDesign, full brief

> One screenshot in, a wall of AI redesigns out. RēDesign fires a UI screenshot at Claude,
> GPT, Gemini, DeepSeek, and Qwen at once, then lines every result up next to the original
> so you can actually compare them instead of pasting the same image into ten chat tabs.

Last updated: 2026-08-23. Canonical source: https://redesign.lunarwerx.com/
Machine-readable pricing: https://redesign.lunarwerx.com/pricing.md

## What it is

RēDesign is a control panel that runs on your own machine (Bun + Hono backend, Vue 3 +
Vite + Tailwind + shadcn-vue front end) and talks only to the AI model APIs you give it
keys for. Drop in a UI screenshot, pick models and prompt presets, hit Run, and every
model's redesign renders as real, self-contained HTML in a locked-down iframe next to the
original, so results can be judged side by side instead of eyeballed one chat tab at a
time. Adding another model is a JSON config entry, not a code change, and everything in
the UI has a CLI twin plus an MCP server so an AI agent can drive it too.

## Who it is for

Developers and designers who want a fast first pass at redesigning an existing screen,
comparing several AI models' takes at once, without hand-building mockups or manually
re-running the same screenshot through five separate chat apps.

## Key facts

- Free to download, MIT licensed, source on GitHub.
- Runs on Windows, Linux, and macOS (Apple silicon). Windows has a standalone GUI
  executable with the web app embedded (no console window), plus an optional
  system-tray build.
- From source: Bun 1.2 or newer. `bun install && npm run build`, `cp .env.example .env`,
  `bun run src/index.ts serve`, then open 127.0.0.1:5178.
- Ships with 10 models by default: Claude, GPT, Gemini, DeepSeek, Qwen, and Meta AI. Claude, GPT,
  Gemini, and Qwen see the screenshot natively; DeepSeek works from an auto-generated
  description of it. Any OpenAI-compatible endpoint can be added as config, no code change.
- 8 prompt presets out of the box: Faithful Refresh, Bold Reimagine, Clarity &
  Accessibility, Conversion, Minimalist, Material 3, Approachable, Playful, or write
  your own.
- Every model fires in parallel on one screenshot; you can also ask one model for several
  variants in the same run instead of just one roll of the dice.
- Every run is grounded: a vision model inventories the screenshot once and hands that
  description to every model alongside the image, so redesigns keep the real content
  instead of dropping a tab or inventing a metric.
- Key management: paste one key or a pile at once, RēDesign works out which provider each
  belongs to and pools them; a failing key gets benched and quietly retried later, so one
  dead key doesn't cost you the whole batch.
- Cost awareness: an estimate before you hit Run, plus a per-run cost meter afterward. A
  `--mock` dry run costs nothing.
- Screenshots and API keys never leave your machine except to the model APIs you chose.
  Keys live in a local `.env` file, owner-only (0600 on Unix), never written into a run's
  saved output, and stripped from provider error text before it's shown.
- Reproducible and scriptable: every run is written to a manifest, with a full CLI twin
  (`bun run src/index.ts run|models|keys|mcp`) and an MCP server for agents.
- Current version: 1.6.6.

## Install

- Packaged builds: download from GitHub Releases, run it, no install step. Windows also
  gets `redesign-windows-x64.exe` (GUI, no console) or a system-tray build.
- From source (any supported OS): Bun 1.2+, `bun install && npm run build`,
  `cp .env.example .env`, `bun run src/index.ts serve`.

## Pricing

RēDesign itself is free and MIT licensed, no subscription, no account, no usage cap from
LunarWerx. The only cost is your own usage of whichever AI provider APIs you configure
(Claude, GPT, Gemini, DeepSeek, Qwen, or any OpenAI-compatible endpoint you add), billed
directly by that provider to your own key. See /pricing.md for the machine-readable
breakdown.

## FAQ

- **Is RēDesign free?** Yes, free to download, MIT licensed, source on GitHub.
- **Which AI models does it support?** Claude, GPT, Gemini, DeepSeek, Qwen, and Meta AI by default,
  plus any model added as a JSON config entry.
- **Can I add my own AI model?** Yes, via JSON config, no code change required.
- **Does it keep my screenshots and API keys private?** Yes, localhost only, keys and
  screenshots never leave your machine except to the APIs you chose.
- **What operating systems does it run on?** Windows, Linux, and macOS on Apple silicon.
- **Can I try it without spending on AI API calls?** Yes, `--mock` costs nothing.
- **Does it work offline?** The control panel runs locally with no cloud account, but
  generating a real redesign needs internet access to reach the model APIs you configured.
- **How is it different from asking ChatGPT or Claude directly?** A single chat gives one
  model's take per conversation; RēDesign fires the same screenshot at ten models in
  parallel and lines every result up as real HTML you can open, not just a chat reply.

## Limitations

- Not a hosted service. It runs on your own machine, so you need Bun (for source builds)
  or a packaged download, plus your own API keys for whichever models you use.
- Generating real redesigns requires internet access to reach the configured model APIs;
  only the local control panel and a `--mock` dry run work without it.
- DeepSeek does not see the screenshot natively. It works from an auto-generated text
  description of it, so its take can miss visual detail the other models catch directly.
- You bear whatever cost each AI provider charges for the API calls RēDesign makes on
  your behalf; RēDesign shows an estimate first but does not cap or subsidize that spend.

## Links

- Home: https://redesign.lunarwerx.com/
- Pricing (machine-readable): https://redesign.lunarwerx.com/pricing.md
- Source code: https://github.com/LunarWerxs/ReDesign
- Download: https://github.com/LunarWerxs/ReDesign/releases/latest
- License: MIT
- Parent studio: https://lunarwerx.com/