Can a prompt replace Lovable?

Build a tool that generates a complete, runnable full-stack starter app from one descriptive prompt — frontend, database schema, and auth already wired together, not just isolated UI components like this catalogue's v0 entry. Use a fixed stack, Next.js, Postgres, a simple email/password auth library, so the model's job is filling in a known template rather than choosing infrastructure. Read a model key from ANTHROPIC_API_KEY or OPENAI_API_KEY. Given a prompt like 'a habit tracker where users log daily habits and see a streak', have the model infer a Postgres schema (users, habits, habit_logs tables with sensible relations), generate CRUD API routes for it, generate matching frontend pages (a dashboard, a habit list, a log-entry form), and wire auth so routes are protected. Write all generated files to a real project directory, run migrations against a connected Postgres database, and start the dev server so the user has something running immediately. Support one follow-up refinement prompt, such as 'also add a weekly summary view', that the model applies as a diff against the existing generated project, rather than only supporting a single one-shot generation. Do not build many-turn iterative refinement, one-click cloud database provisioning, or visual UI editing — those are out of scope; this generates a solid starting scaffold plus one refinement, not an ongoing conversational IDE. Requires an Anthropic or OpenAI API key, hosting, and a Postgres database.