For a complete explanation of the product, current MVP, and generation pipeline, read the Vaulter master document.
Answer 5 plain-English questions about your app idea. Get back a real, grounded Build Contract written live by an LLM, not a static template, that you can paste into any AI coding tool to start building correctly from the first prompt.
Vibe coders skip invariants, tenant isolation, and failure-mode thinking not because they don't care, but because nobody hands them the questions before they start prompting. Vaulter Studio is that front door: five concrete questions, each with a real architectural consequence (RLS vs. caching, WebSockets vs. polling, object storage vs. none, payment/email integrations vs. none, offline-first sync vs. none), assembled into a brief an LLM writes fresh for your specific answers.
The tech-stack grid and file tree on the right update per answer. Nothing renders unless a specific choice actually triggered it. The final document is generated by a real OpenAI call (app/api/generate/route.ts) with a system prompt that explicitly forbids inventing technology outside what your answers produced, forbids padding sections with generic boilerplate, and forbids listing mutually-exclusive choices (e.g. never both a WebSocket store and a polling library for the same real-time-vs-refresh decision).
Next.js 16 (App Router, TypeScript), Tailwind v4, framer-motion, OpenAI SDK.
npm install
cp .env.example .env.local # add your OPENAI_API_KEY
npm run devRequires an OpenAI API key with billing enabled (OPENAI_API_KEY) and a model string (OPENAI_MODEL, defaults to gpt-5).