Framer 3.0: What's New (and What Doesn't Change When You Export)

Framer 3.0 launched June 16, 2026, and hit #1 on Product Hunt the same day. The headline additions are AI Agents (built into the canvas — they design pages, edit components, manage CMS content, and can fix site issues), Branching (a safe, separate version of your site you can test and review before merging into production), On-Page Editing (editing a live page directly in the browser rather than only in the canvas), a refreshed Community, and new pricing.

If you're planning to export a Framer 3.0 site — to this tool or any other — here's the honest answer to the question that actually matters: does any of this change what comes out the other side?

What doesn't change

An export tool (this one included) works from the published site — the actual HTML, CSS, and JS Framer's hosting serves to a visitor. It has no visibility into, and no dependency on, how that markup was produced inside Framer's editor.

That means:

• AI Agents generate pages using Framer's normal components and layout primitives. The DOM an agent-built page publishes is structurally the same kind of DOM a hand-built page publishes — same class conventions, same runtime hooks. There's no separate "AI-generated" markup dialect to account for.
• Branching is entirely pre-publish. It's how you manage which version becomes live; once you merge and publish, the result is an ordinary published Framer page like any other. Nothing about having used a branch survives into the exported output.
• On-Page Editing changes where you make an edit (in the browser, against the live page) rather than what the resulting published markup looks like. The published result is the same shape as an edit made the traditional way in the canvas.

Where to actually be careful

The one place this is worth a real caveat: AI Agents are new enough that if they start generating layout patterns Framer's editor has never produced before — a genuinely novel component structure, not just a new arrangement of existing ones — an export tool's detection logic (for things like nav structure, image handling, or animation start-states) could hit an edge case nobody's specifically tested yet. We haven't seen that happen, but "haven't seen it" isn't the same claim as "verified against every possible AI Agent output," and we'd rather say that plainly than pretend otherwise.

The practical answer: export a Framer 3.0 site the same way you'd export any other — the Next.js export for a project split into named section components, Framer to HTML for static files, or an Astro project, all three keeping Framer's runtime — and treat the preview step as your actual verification, not a formality. That's true for a 3.0 site exactly as much as it was for 2.x.

The bigger picture

Framer's own positioning is now explicitly "AI website builder for professional sites" — design, CMS, hosting, SEO, and now agents and branching, all in one stack. That's a reasonable pitch for staying inside Framer. It's a separate question from whether you own the result once it's built, which is the question this tool exists to answer: yes, at any point, in a form you can self-host — though it still loads Framer's runtime and fonts from Framer's CDN, so the Framer project stays published.