Prototypes that develop your PRD, not just transcribe it

Generated prototypes were coming out too thin — a literal, skeletal transcription of the PRD instead of something that feels like a shipped product. This release rebuilds the engine so a prototype develops your spec the way Lovable or Stitch would, while staying a clickable, multi-screen HTML/CSS deliverable you can put in front of a stakeholder.

A reasoning model does the building

The per-page HTML authoring now runs on the Pro reasoning tier instead of the lightweight model that drafted earlier prototypes. Pages come out with deeper hierarchy, real layout sophistication, and far more developed content. Iterative chat edits stay on the fast model, so refining a prototype is still quick and cheap.

A shared component kit, so every screen matches

Thinkr now owns the design vocabulary. A token-built component kit — buttons, cards, fields, tables, badges, stats, empty and loading states — is baked into every page, and the model composes screens from it instead of hand-styling each one from scratch. The result: a prototype's pages look like one coherent product, not a dozen independently-styled screens.

It fills screens in — and marks what's illustrative

Earlier prototypes refused to add anything the PRD didn't spell out, which left screens sparse. Now the generator fleshes out plausible sample content — realistic rows, filled and empty states, believable copy — so a screen feels complete. Anything invented that could be mistaken for real data is visibly marked as a sample, so the honesty rules still hold: no fabricated metrics, testimonials, or logos presented as real.

Match a reference screenshot

Upload a reference image when you create a prototype and the build will track its visual language — palette, type pairing, density, spacing — while staying within your design tokens. A fast way to point at a look you like and get something in its spirit.

Every page critiques and repairs itself

After a page is generated, deterministic checks run over it — broken links, missing primary CTAs, accessibility basics, viewport overflow, whether it actually used the component kit — and a second reasoning pass elevates the page and fixes what the checks found. Anything that still doesn't pass is surfaced to you as a quality note in the page index.

Build straight through, or review the plan first

By default, Thinkr now plans the page structure and builds straight through — pick a PRD, hit Build prototype, and watch the pages stream in. Prefer to steer the structure first? Turn on Review the page plan before building and you get the page list and CTA wiring to edit before anything is generated.