UnBuild

Reverse engineer anything.

Paste a GitHub repo, website URL, product page, API spec, or describe any idea in plain text. UnBuild deep-analyzes it and produces everything an AI coding tool needs to rebuild it from scratch.

Live at unbuild.tech

What you get

Every analysis produces four deliverables:

Output	What it is
Implementation Blueprint	Multi-phase build plan with parallel agent decomposition, data models, API endpoints, and validation steps. 2000-4000 words.
CLAUDE.md	Production-ready AI agent config — design system, dev commands, component specs, and product-specific gotchas. 800-2000 words.
Initial Build Prompt	The single prompt you paste into any AI tool (Claude Code, Cursor, Codex, Windsurf). Tells the AI to research, plan, build, and test. 800-1100 words.
Architecture Analysis	Evidence-based JSON — detected tech, component graph, user journeys, reproduction challenges.

What you can analyze

• GitHub repos — Full codebase analysis with file tree, dependencies, architecture patterns
• Websites — Renders with Playwright (real browser), extracts design system, infers backend
• Product pages — Component teardown, bill of materials, open-source alternatives
• API specs — Endpoint catalog, schema definitions, integration guide
• Plain text ideas — Describe anything, get a researched blueprint with tech stack and build phases

Quick start

Web

Visit unbuild.tech, paste your input, get results.

Every analysis creates a shareable link (e.g., unbuild.tech/r/craigslist-org) you can revisit or share with your AI coding tool.

CLI

npx unbuild-dev vercel/next.js
npx unbuild-dev https://stripe.com
npx unbuild-dev --help

Self-hosting

git clone https://github.com/detreaba/unbuild
cd unbuild
pnpm install
cp .env.example .env.local
# Add your API key to .env.local
npx playwright install chromium
pnpm dev

How it works

1. Fetch & Extract — Uses Playwright (real Chrome browser) to render pages, read source code, and extract CSS/HTML structure
2. Understand the Product — Chain of Verification ensures every claim is evidence-based. Identifies what the product DOES, not just what it looks like
3. Architect the Build — Three separate LLM passes: architecture analysis (JSON) → implementation blueprint → CLAUDE.md
4. Generate the Prompt — Dedicated LLM call produces the Initial Build Prompt, informed by both the blueprint and CLAUDE.md

Admin panel

Self-hosted instances include a full admin panel at /admin:

• API Keys — Add/test/rotate keys for OpenAI, Anthropic, OpenRouter, Google, Ollama
• Model Selection — Choose model, temperature, max tokens
• History — Every analysis logged with timing, status, and model used
• Rate Limiting — Per-IP throttling with allowlist
• Analytics — Usage stats, referrers, error rates

LLM providers

UnBuild works with any LLM provider. Set one in .env.local:

Provider	Env var	Default model
OpenAI	OPENAI_API_KEY	gpt-4.1
Anthropic	ANTHROPIC_API_KEY	claude-sonnet-4
OpenRouter	OPENROUTER_API_KEY	anthropic/claude-sonnet-4
Google	GOOGLE_GENERATIVE_AI_API_KEY	gemini-2.5-pro
Ollama (local)	OLLAMA_URL	llama3.1

No API key? Falls back to Claude Code CLI (uses your subscription).

Environment variables

# LLM (at least one required)
OPENAI_API_KEY=
ANTHROPIC_API_KEY=
OPENROUTER_API_KEY=

# GitHub (recommended — 60 → 5000 req/hr)
GITHUB_TOKEN=

# Admin panel
ADMIN_PASSWORD=change-me

# Optional
OLLAMA_URL=http://localhost:11434
OLLAMA_MODEL=llama3.1

Tech stack

• Framework: Next.js 16 (App Router) + React 19 + TypeScript
• Styling: Tailwind CSS 4 + Framer Motion
• Browser automation: Playwright (renders JS-heavy sites)
• LLM: Multi-provider abstraction (6 providers)
• Auth: JWT sessions with password-based admin login
• Storage: JSON file database (no external DB required)
• Testing: Vitest (57 unit tests)

Project structure

src/
  app/
    api/
      analyze/            — GitHub repo analysis pipeline
      analyze-website/    — Website analysis pipeline (Playwright)
      analyze-universal/  — Product, API, text analysis pipeline
      admin/              — Admin API (keys, config, history, analytics)
      results/[slug]/     — Shareable result API
    admin/                — Admin panel UI
    r/[slug]/             — Shareable result pages
  components/
    sections/             — Landing page sections
    tool/                 — Analysis tool UI (input, loading, results)
    layout/               — Navbar, footer
    ui/                   — Reusable components
  lib/
    analyzers/            — Input detection, tree analysis, website/product parsing
    generators/           — Multi-pass LLM blueprint generation
    llm/                  — Provider abstraction + system prompts
    admin/                — Auth, data store, types
cli/                      — Standalone CLI tool

Contributing

Contributions welcome. Open an issue first to discuss what you'd like to change.

License

MIT