Blueprint-driven website generation
The tool that builds your site
shouldn't ship you a template.
FoundryKit turns structured product intent into a blueprint — classification, a design gate, hostability, and proof rules — then generates production-grade source you own. Every build is gated against generic, templated design before it ships.
I need a modern website for a mobile mechanic in Pittsburgh. Customers should be able to see services, trust me quickly, and request a quote. Make it feel tough, reliable, and local — not like a generic repair shop template.- CLASSIFYlocal-service
Audience, offer, primary action, and output target — derived before any layout.
- DESIGN GATElocal-trust-path · card budget 1
Composition + avoid-list per site type. No abstract hero, no decorative card grid.
- HOSTABILITYstatic · domain-ready
Hosting model, recommended target, and blockers, decided up front.
- PROOFverified reviews only
Fabricated testimonials and unsupported metrics removed automatically.
Real outputs
One prompt, many realities — not one skeleton.
The design gate forces each site type into its own composition. Same engine, genuinely different sites.
Local service
Repair-shop trust flow
- Layout
- black/yellow commercial hero + reviews + service strip
- Motion
- booking path and service reveal
- Proof
- reviews, services, quote CTA
Fitness / wellness
Local gym startup
- Layout
- full-bleed photo hero + values triptych + signup band
- Motion
- image-led scroll pacing
- Proof
- classes, trial form, community sections
Learning product
Workspace proof surface
- Layout
- course overview + task rail + editor + output panel
- Motion
- run/check/proof interaction loop
- Proof
- visible editor, output, Check Work, reflection
Product / ecommerce
Handmade product commerce
- Layout
- product-photography hero + detail module + cart path
- Motion
- finish choice and cart microinteraction
- Proof
- product photography, price, quantity, add-to-cart, shipping note
Why the output isn't generic
The intelligence is in the gate, not the prompt.
Every build passes a machine-readable design gate, a hostability assessment, and a trust-theater check before it ships. Card-walls past budget get flagged; fabricated testimonials and unsupported metrics get removed. That is why the output doesn't read as AI.
- Design gatecomposition + card budget per site type
- Hostabilitystatic / runtime, domain-ready, blockers
- Proof rulesverified-only — no trust theater
You own the output
It ends in ordinary files you can host anywhere.
No lock-in. The export is plain source plus a HOSTING.md for your recommended target.