Your agent operates the website. Sitehold is what it operates.
Sitehold is the infrastructure underneath an agent-run website: the interfaces an AI agent uses to inspect a site, change it, publish it and keep it correct over time. You describe outcomes. The agent does the operating. Sitehold makes that reliable.
Not a website builder. Not a chatbot. Not a one-shot generator. The layer in between.
“We shipped archive uploads today. Add them to the changelog.”
Finds where the changelog lives, writes the entry, decides it can deploy without asking you.
A website is never finished. Sitehold is built for everything after the first publish.
Generators stop when the site exists. Real sites keep changing: hours, locations, products, events, mistakes. Sitehold keeps a site in a state an agent can return to, inspect and change again — for as long as the site exists.
- Once Create A site is held. It has structure, content and a domain.
- Step 1 Inspect The agent reads what exists before touching it.
- Step 2 Change Content, pages, navigation — as a recorded change.
- Step 3 Publish The change becomes the live website, or waits for you.
- Step 4 Verify The agent checks the live site did what was intended.
- Step 5 → 1 Maintain Weeks later, the same agent comes back and starts again at Inspect.
One request, followed all the way to the live site.
Pick a request. The left side is what the agent does and which Sitehold operation it uses. The right side is what happens to the website.
Five jobs. Every operation belongs to one of them.
Sitehold's interface is organised around what an agent is trying to do to a site, not around features. Operation names below are representative of the MCP tools and API calls an agent discovers.