Check. Install. Ask. Approve.
The check is free and runs from outside. Everything after it happens on your site, by an agent you can watch and undo. Seven steps, and what the agent says when a step is not its to take.
The free check reads your site from outside
Enter a domain and an address. The 23 checks run on your public pages in minutes with nothing installed: can engines reach the pages, read them, know who you are, and find others vouching for you. Every finding shows the evidence and what the agent would do about it.
free · no account needed for the reportCreate a free account and keep the site
A free account keeps one site and its checks, five a month. You can stop here and fix things by hand with the guides; nothing is installed yet.
free · one sitePut the agent on the site
WordPress: upload the plugin, activate, paste the token. Coded site: one file in your middleware, Worker or Edge Function with the token in an environment variable. Repository: install the Reflexa GitHub App on that one repo. Then press "Check the connection": the agent proves it is on your real domain.
Starter or Growth · one token per siteSwitch on what the check found
Findings the agent can fix itself show a button. Allow AI crawlers in robots.txt, add Organization data, turn the sitemap on, publish /llms.txt: one click each, applied from inside the site, reversible.
your clickAsk for a goal in your words
"Get the pricing page read by AI." "Why did AI stop opening the blog?" The agent builds the facts for free, runs one model pass, and returns a diagnosis and a plan. Each item carries its verdict: the agent switches it on, the agent writes it and you approve, or a person has to do it.
one priced pass, shown before it runsApprove the exact words
Text items show the proposed title, description, opening or FAQ in full. Approve one or all. What you approve is published as a revision; what you do not approve is never touched.
your clickThe agent confirms it landed, and keeps watching
After every change it re-reads the live page and reports landed, not yet, or failed. A weekly check runs by itself, AI visits are counted every day, and the next goal starts from what actually moved, never from an assertion.
weekly check · AI traffic · undo at any timeWhat the agent says when it cannot
Before any diagnosis, the agent answers whether the thing you asked is within its hands. The five verdicts are the product's honesty, in code.
| Verdict | Meaning |
|---|---|
| now | A switch it can apply from inside the site; your click applies it. |
| with your click | Text it writes and publishes once you approve the exact words. |
| proposal only | Text it writes for you to paste, where it cannot run inside the site. |
| blocked | Something it cannot do until one step is taken, and which step. |
| not mine | Outside its hands entirely, and who does it instead. |
Questions people ask
How long does the free check take?
A few minutes for a typical site. The report opens by its own link and the result follows by email.
What does a goal cost?
The facts are free: checks, AI visits, Search Console, page status. One model pass writes the diagnosis and the plan, and the price is shown before it runs. Free accounts run no goals; Starter includes around 200 answers a month, Growth around 1,000.
What if the agent cannot do what I ask?
It says so before the diagnosis, with a verdict: now, with your click, proposal only, blocked (and the one step that unblocks it), or not mine (and who does it instead). It remembers the verdict so it does not promise it again.
How do I know a change actually landed?
The agent re-reads the live page after every change and looks for its own marker. It reports landed, not yet (a deploy has not run), or failed. "Applied" is never taken on trust.
Start with the free check. Then tell the agent what you need.
The free check runs the 23 checks on your public pages. Nothing is installed, nothing changes until you press it.
Free check from outside in minutes · the report by email · nothing installed yet