No AI products today. This page is a public preview of what we're planning — a Webflow-native AI agent trained on our embed library and design patterns. Ships after Hublumen and Baselumen.
An AI co-builder is only as good as the patterns it's trained on. Hublumen and Baselumen are how we're encoding those patterns into a structured library — Stacklumen AI inherits everything they ship.
Component library application — the structured catalog of Webflow embeds and React patterns that the AI will be trained on.
Client portal application — feedback, approvals, and project state that gives the AI context about real-world use cases.
Webflow-native AI co-builder, trained on our patterns and the Hublumen library. Generation, debugging, translation — scoped to a stack we know works.
A short list of capabilities we're scoping for the first release. Some are concrete (we know how we'd build them); some are concepts (we want them, the approach is open). Status reflects that.
Translate a Figma frame or natural-language description into a Webflow embed that respects the 50KB limit. Output uses the Hublumen pattern library, so the code is consistent with our handcrafted work.
Describe a scroll-triggered sequence in plain English; get a working GSAP timeline back, scoped and minified. Built on our reference patterns from real client work.
Paste your Webflow custom code. Get back a list of class collisions with Tailwind utilities, Radix primitives, or third-party scripts — with rename suggestions.
Paste an embed that's over the 50KB limit. Get back a refactored version that fits — with a diff explaining every change.
Generate React + TypeScript components matching the Hublumen design system from a description. Approach is still open — we may use codegen, model fine-tuning, or both.
Generate Vitest or Playwright suites for components and embeds. Approach depends on how reliable model-generated tests prove to be in real codebases — we're researching.
The market doesn't need another general-purpose code AI. It needs tools that are honest about scope and excellent within it. Here's the approach.
The model is fine-tuned on Hublumen patterns and our agency codebase — not the entire internet. Narrower scope, sharper output.
Outputs are diffable, attributable to source patterns, and editable. You see why the model made each decision before you accept it.
Webflow custom code is a niche where most AI tools produce garbage. We've shipped 40+ sites in it — that's the asymmetric advantage.
First release covers two or three concrete use cases that work well, not ten that work poorly. We grow scope based on what people actually use.
The waitlist is the honest version of "early access." We email you once when there's something to actually try, not before.