Let AI scale your website,
without losing control.

An AI-powered website is a site whose day-to-day changes are produced by AI coding agents, like Claude Code, Codex, or Gemini, working directly against the site's real codebase, instead of by hand through a CMS. The tools to build and run a website this way already exist; Dispatch makes them manageable for your organization by adding the visibility, context, approvals, and history around every change.

It is safe when the same controls a CMS provided still exist: every change previewed before it ships, approvals owned by named roles, AI-authored work attributed to the agent that made it, and a restore path back to any prior version. Git-based sites make each of these stronger than most CMS workflows. What makes AI unsafe is not the agent; it is running one with no management layer around it.

Traditional CMS platforms solved a real problem: structure, permissions, publishing controls, and a safe interface for non-technical teams. AI changes how the work is produced: an agent can create pages, improve content, and make sitewide changes directly against modern website code. Dispatch preserves the CMS's safety net (visibility, approvals, roles, history) without constraining what AI can do. Keep the safety net; remove the dependency on the CMS.

Connect your site's GitHub repository and your hosting platform (Vercel, Netlify, or Cloudflare) from the Websites section. Dispatch syncs your pages, scores each one for AEO readiness, and starts recording every commit, pull request, and deployment in the activity feed with attribution. From there, add your brand context and set roles so approvals are enforced from the first change.

No. Your website code, content, and history live in your repository and host, not in Dispatch; turn Dispatch off and the site keeps running exactly as it is. Everything you create inside Dispatch, prompts, context, agents, workflows, documents, and the full request and approval record, exports from Settings as JSON and Markdown at any time.

Yes. Every change request gets its own branch and preview, so people and agents work in parallel without overwriting each other. Roles decide who can approve, high-risk requests need two distinct approvers, and every action lands in one activity feed with the person or agent attributed. The Inbox shows the whole team what needs them, and Slack alerts carry the draft when someone is needed.

The request. Every job an agent gets from Dispatch is scoped to a page, a risk tier, and your approved context, and the agent works on a branch, never on production. Your team reviews the preview before anything ships, a second approver is required for high-risk work, and one click restores the previous version if something still slips through.