AI agents

More and more people ask an AI agent to do things on the web for them. AI agents in Flamio On-Live prepares your site for those agents: they find its sections and forms, prepare a form for the visitor, and hand the visitor a link to it. You see what agents looked for, where they went and where they got stuck. Visitors still send every form themselves.

What you get

• Your site as tools for agents. Sections you tag become places agents can link to; forms you tag become actions agents can prepare.
• Control over what agents fill. Each form field is filled by the agent, by the visitor only, or hidden from agents.
• A view of what agents will find. A scan lists every page, section, form and field agents get, and the markup issues to fix.
• One switch to publish or withdraw your site from agents.
• A signed-in part kept for signed-in visitors, when you want it.
• Agent activity: what agents searched for, the actions they asked for, where they were refused, which agents came and from where.

How it works at a glance

1. Set up and verify. Add your site under AI agents and prove you own it with a tag on your home page.
2. Tag your markup. Add attributes to the sections and forms agents should find.
3. Scan. Flamio reads your pages and shows the site tree agents will get.
4. Publish. Turn on Agents can use this site.
5. Agents find your site. Agents that open your site in a browser with WebMCP see its sections and tagged forms as tools.
6. Watch the activity. See what agents did and fix what blocked them.

AI agents lives in the sidebar under Flamio On-Live > AI agents, and is set up per project.

What works today

Part of AI agents runs in the Flamio web SDK on your pages, from version 0.11.0. That version is not released yet: npm serves 0.9.0, which has none of these parts. Until 0.11.0 is out:

• A form link an agent hands out opens your page, but the form does not show the values the agent proposed.
• Only a browser that reads tagged forms on its own, such as Chrome with the WebMCP origin trial token on your pages, gets them as tools. Sections reach browser agents only through the SDK. See How AI agents find your site.
• Signed-in visitors, and the Programs and Actions done counts on Activity, also need the @flamioai/agent-panel package, which is not released either.

Setup and domain verification, markup, scans, publishing and your site's MCP address work today.

FlamioSiteReader in your logs

To read your site, Flamio uses its own crawler. Every request it makes carries this User-Agent:

FlamioSiteReader/1.0 (+https://flamio.org/agents)

• When it visits. Only after you verify your domain. Before that, Flamio requests your home page only when you check the tag. After verification it visits right after you verify, when you click Scan now, when you publish, and once a day while the site stays published.
• What it reads. Public pages on your verified site's own host, found through your sitemap and the links between pages. It requests the page HTML and does not run the page's JavaScript.
• What it never does. It never submits a form, never signs in and never clicks anything.
• How to stop it. Unpublish the site on AI agents > Publish. Removing the verification tag also takes the site away from agents once the next checks fail.

Read more on the FlamioSiteReader page.

Limitations

• Flamio does not submit a form for the visitor — instead, an agent prepares a tagged form, proposes values for its Agent fields and hands the visitor a link; the visitor reviews the form and sends it.
• Flamio does not find forms your page builds only with JavaScript — its scan reads the HTML your server sends, so such a form is missing from Site tree, from agent search and from the links agents hand out. Browser agents with WebMCP still see it once it appears on the page. Instead, render the form on the server, then click Scan now on Site tree to confirm it is found.
• Flamio does not let agents fill your visitors' email, phone or card details — Visitor fields are typed by the visitor and Secret fields are never listed to agents. A browser's own agent (Chrome WebMCP) is the exception: it may read every named field of a tagged form, including email, because browsers ignore the role hints.

Next steps