How it works
Two paths to a paste-ready spec
One uses retrieval-grounded synthesis. The other walks you through a constraints-aware wizard. Both produce a spec you paste straight into Claude Code, Cursor, Codex, Windsurf, Antigravity, Copilot, ChatGPT, or any AGENTS.md-compatible tool.
RAG demo
Two fields. Pick a tool, paste a project brief. A hybrid retrieval pipeline pulls the most relevant chunks from a curated YAML knowledge base; an LLM stitches them into a meta-prompt with hoverable citations.
Open the RAG demo →Guided wizard
Around ten questions about your stack, budget, scale, and AI tool. A rules engine flags conflicts before the plan is written, then a validator hard-fails on any missing section.
Open the wizard →Six checks the rules engine runs before any plan is written
Applies to both paths. The wizard surfaces these inline; the RAG demo grounds its citations against the same engine.
- Stack picked against your constraints. Budget, scale, compliance, AI tool. Fed to a rules engine that recommends for your situation, not a generic starting point.
- Conflicts flagged before you build. Native iOS on Windows-only, EU users on a US-only host, a paid payment processor on a £0 budget. The conflict and the fix get named before you waste the weekend.
- Structured output, every time. Phase 0, Daily Workflow, Troubleshooting, Launch Checklist. A validator hard-fails on any missing section.
- Format tailored to your AI tool. Claude Code gets CLAUDE.md + TASKS.md. Cursor gets workspace rules. Same plan, the right format.
- Pinned library versions. Tell us “Next 15.1.3, Python 3.12”. The prompt locks your AI to those versions instead of reaching for stale training data.
- Edit one answer, see what shifts. Flip £0 to £20/mo. Database, hosting, payments re-recommend instantly. No restart from scratch.