A site for people, and for the agents acting on their behalf
The pattern of search, click ten links, read is giving way to ask, get an answer, maybe verify. Websites are starting to serve two kinds of visitor: people on screens, and software that reads, compares, books and buys on a person's behalf. Here is what a WPExit Rebuild does about that, with no vapor.
Shipping today
Readable by machines
Every page is published twice: as HTML for people and as Markdown at the same URL plus index.md. An llms.txt index lists every page with a one-line description, and llms-full.txt carries the whole site in one file. schema.org data describes the business (LocalBusiness or Organization), each page, and which parts should be spoken aloud by voice assistants. Headings, landmarks and alt text are checked on every build, which is what makes a page legible to a model as well as to a screen reader.
Act, don't scrape: a hosted MCP server per site
Model Context Protocol is how AI applications connect to tools. Each rebuilt site gets its own endpoint, advertised in the site's agent manifest and llms.txt. An agent that finds it can call site_info, list_pages, get_page, search, list_forms and submit_form: read your pages as clean Markdown, find the one that answers the question, and send a contact or quote request through your real form, which notifies you by email exactly as a visitor's would. No HTML parsing, no guessing, and your AI access policy is the first thing it is told.
Verifiable: every build is signed
When synthetic text and images are everywhere, the useful question stops being is this real and becomes who published this, and has it changed. Every Rebuild publishes /.well-known/provenance.json: the SHA-256 of every page and text file, signed with an Ed25519 key that is unique to your site, with the public key beside it. Anyone, human or agent, can fetch a page, hash it, and confirm it is the version you published. The key stays with us encrypted; only its public half ships.
Your AI access policy
You choose how AI systems may use your content: open, attribution required (answers must link to the page used), or closed (the known AI crawlers are disallowed). The choice is published in three places that machines actually read: robots.txt, llms.txt and the agent manifest. Search engines are always welcome.
Fast on any connection
A rebuilt home page is typically under 150 KB in under 10 requests, measured on 22 sites. That matters on satellite internet and old phones, and it matters to agents, which pay for every byte they read.
Yours to take anywhere
There is no database and no proprietary format. Your site is a folder of HTML, Markdown and images that any static host on earth can serve. Download it from the portal at any time. Social profile links carry rel="me", so identity verification on the open social web (Mastodon and friends) works from your own domain.
On the road map
- Content credentials for images. C2PA manifests attached to the photos you publish, so an image can carry who published it and when.
- Paid and metered access. A way to let AI crawlers in for a price or a license, for sites whose content is the product.
- Richer actions. Bookings, availability and structured offers exposed as MCP tools, for the businesses that have them.
- Portable identity. One-click verification of your domain for Bluesky and the fediverse.
What this is not
It is not a chatbot bolted onto your site, and it does not send your content anywhere. It is your site, built so that the software now visiting it can understand it, act through it, and prove it came from you. Freeze keeps your site exactly as it is; Rebuild is where the above lives.