How Brease works with your site
The site stays a normal repository. Brease edits it in a sandbox and serves its content through an API.
Brease doesn't host your site and doesn't own its templates. Your site is an ordinary repository on GitHub, deployed by Vercel or Cloudflare, that reads its content from Brease with brease-client.
The parts
- Your repository. Code, components, styles. Brease commits to it only when someone publishes a code change from the editor (or merges a draft's pull request).
- Content. Pages, navigations, redirects and SEO, stored in Brease as immutable releases. Your site fetches them from the delivery API with a read-only token.
- The editor. When someone opens a site in Brease, it starts a sandbox: a fresh checkout of the production branch, dependencies installed, the dev server running, and the live content written to a
.brease/folder. The user sees that dev server as the preview. The AI agent and the user's own edits change the code and the.brease/files there. - Publishing. A content-only change becomes a new release and your site is told to refresh. A change that touches code is committed, checked by your CI, pushed to the production branch and deployed by your host. See Publishing and caching.
What you provide
- A repository connected through the Brease GitHub App, and a Vercel or Cloudflare project for it. See Connect a site.
- Code that reads content with
brease-clientinstead of hard-coding it. - Editor markers (
data-breaseattributes) on the elements that render content, so users can click them in the preview. - A
BREASE.mdthat tells the AI how the site is built. - Optionally a
content.schema.jsonthat describes each page template's fields.
Brease counts a site as integrated when its package.json depends on brease-client and the repository has a BREASE.md. A site that isn't integrated yet can still be opened in the editor: the AI agent can do the integration work, moving hard-coded texts into content and adding markers.