Most small businesses do not need a website project
They need a website. The difference is six weeks, three rounds of feedback, and a folder of screenshots nobody looks at again. Framewise exists to collapse that into a sentence and an afternoon of editing.
The blank page is the bottleneck
Drag-and-drop builders solved the wrong half of the problem. Moving a box is easy. Knowing that a service business needs a services grid above the fold, social proof before the pricing, and a contact block that does not bury the phone number — that is the part that takes experience.
So people buy a template instead, and spend a weekend deleting placeholder content for a company that does not exist, in a layout designed for a company that is not theirs. Or they pay someone, wait six weeks, and end up unable to change a phone number without opening a ticket.
None of those is a tooling problem. They are all the same problem: nobody makes the first draft.
Make the draft, then get out of the way
Framewise reads a sentence about your business and designs a complete site for it — the pages you need, how each one should be structured, in a sensible reading order, with every word written for what you actually do.
Then it stops. Everything it produced is editable, every section can be swapped for another layout without losing your words, and the whole thing can be exported as a codebase you own. The AI writes the first draft; you own everything after that.
That is the entire product thesis. The design system, the override model and the export pipeline are all just what it takes to make that sentence true.
What we decided, and why
These are not values on a wall. Each one is a decision that cost us something, written down so the next hard call gets made the same way.
A finished draft beats an empty canvas
Most website tools hand you a blank page and a hundred decisions. People do not stall because they lack a page builder; they stall because they do not know what the first section should be. So the product answers that first and lets you argue with the answer.
You should be able to leave
Every site is exportable as a real Next.js project with no dependency on us, on every plan including the free one. A builder you cannot leave is a subscription with a design tool attached, and we would rather earn the renewal.
Constraints are a feature
Unlimited CSS control produces broken layouts; a rigid template produces sites that all look alike. Giving the model a bounded design space, and you free-form control where it matters, is the narrow path between the two — and it is where most of our effort goes.
Fast is not a setting
Published pages ship no JavaScript bundle. Interactivity comes from native HTML elements and animation from CSS keyframes. You cannot bolt performance on later; you get it by refusing to add the thing that costs it.
Degrade, do not collapse
The generator never throws. A bad model response, an unknown section, an outright outage — all of it resolves to a working site plus a note about what was substituted. Every integration checks its own configuration and fails alone.
The database decides who owns what
Access is enforced by row-level security on every query rather than by application code remembering to filter. Even the admin panel reads through it, so a bug in our code cannot quietly hand one person another person’s project.
Four decisions the whole thing rests on
Written plainly because the people who care about this tend to care a lot, and the people who do not can safely skip it.
One contract, four consumers
The canvas, the generator, the publish pipeline and the code exporter all read the identical document shape. There is no second implementation to drift, which is why what you export looks like what you edited.
Rendering happens in the browser
Sites are rendered by mounting an offscreen React root and reading the markup back out, because the App Router refuses server-side rendering imports for these components. We tested the obvious approach first; it does not work.
A separate stylesheet for published sites
Published CSS is its own build. Our marketing and editor styles are physically incapable of ending up in a site you ship, which is the only way to be sure they never do.
Section ids are minted locally
Never by the model. A retried completion cannot produce two sections claiming the same identity, so the failure mode that would corrupt a document simply cannot occur.
A short list, kept short deliberately
Promises are cheap and easy to write. These are the ones we are willing to be held to, which is why there are only five of them.
If you catch us breaking one, say so.
- No fake testimonials or invented customer logos on this site.
- No dark patterns in the cancellation flow.
- No selling or sharing of what you write in the idea box.
- No holding your work hostage behind a plan.
- No shipping a tracking bundle to your visitors.
Early access, and honest about it
Framewise is a small team building in the open. The generator, the editor, publishing and the GitHub export all work today — that is what the feature pages describe, and none of it is aspirational.
What is still moving: pricing is not final, some integrations are newer than others, and what the model can design keeps widening. We would rather tell you that than dress an early product up as a finished one.
- The product works end to end today
- Pricing is early-access and will settle
- Bugs get answered by the people who wrote them
Talk to us directly
At this size, feedback goes straight to the people who can act on it. If something is broken, confusing, or nearly right but not quite — that is the most useful thing you can send us.
Get in touchThe fastest way to understand it is to use it
No account, no card. Describe what you do and see what comes back.