Hi Product Hunt, I'm Zaneta and I work in marketing at Pagepro. Our CTO Jakub helps me with the technical questions.

What happens when your marketing team asks an AI agent for a new landing page?

It builds one fast. But the agent doesn't know your design system, your blocks or your Sanity schema, so it guesses. In a Next.js + Sanity site one new section touches 7 places: the schema, the query, the component, a gallery fixture, the Markdown serializer, the insert menu and a tile. Miss one and the page still renders, but editors can't find the section.

We build marketing sites on Next.js and Sanity for clients, and Nexity is the setup we used in 20+ projects. It's open source under MIT.

It gives the agent the rules before it starts. AGENTS.md says how to build and what to ask first. DESIGN.md holds the look. BRAND.md holds the facts and numbers the copy must use. The create-section skill checks whether a section already exists before it builds a new one. Four read-only agents then review the result, so the agent that wrote the code doesn't grade its own work.

For the people who run the site after launch, it's a normal Sanity Studio at /admin with about two dozen sections, click-to-edit preview and redirects they manage themselves. There are also skills for marketers (brand kits, source checks, blog post reviews) that work on any site and any CMS.

It only fits Next.js + Sanity sites. If that's your stack, you need a free Sanity account and one command: npm create sanity@latest -- --template Pagepro/nexity/base. The command sets up the site. The skills and reviewer agents live in .agents/ in the repo, and you copy them into your project.

Repo (MIT): github.com/Pagepro/nexity

Live demo, no account: meridian.nexity.build

If you're a founder or CTO whose team already builds with agents: what would you want an agent to know about your site before it touches it?