Privacy — the readiness scorer
Last updated 7 October 2026.
Who holds it
Your data is held by TrainedBooks, an independent software business in California, United States, which operates the agentwares products. Questions and deletions: https://agentwares.vercel.app/contact.
Data is stored in the United States. If you are in the EEA or the UK, that is a transfer outside your region, and using the tool is you asking us to provide it from there.
No account, no cookies
The readiness scorer needs no account, key or signup. It sets no cookie, stores nothing in your browser and does not follow you to other sites.
When an AI assistant calls it as an MCP tool, it receives the arguments of that call and nothing else: not your conversation, your files or your account with the assistant.
What it counts
Its calls are not counted. Its pages are:
- Each page you open here sends one page view: this site's host and the page's path (never its query string), the host of the page that linked you here if there was one, and two yes-or-no flags — whether the browser looks like a crawler and whether it is one of ours. The browser's user-agent is read to set those flags and then discarded. No cookie, no fingerprint, no network address.
What it keeps about your request
- The scan you ask for: the URL, what the browser saw on that site's public pages (addresses, page titles, the form fields it found, what each step returned), the score, and the model's reading of the site's terms and docs. It is published on a public results page and as a badge — that is the point of the tool.
- A hash of the network address that asked for the scan, stored with it to enforce the limit of scans per client per hour. The hash is made with a random salt that changes every UTC day, and a day's salt is deleted as soon as a scan is requested on a later day; from then on nobody, including us, can tell which address made it. The address itself is never stored. Scans requested before the salt was added, in October 2026, carry an older hash made without one.
- Nothing at all for a domain that has not proved control: the ownership token is computed when asked for, not stored.
If you write to us through the contact form, that message is covered by the privacy note at https://agentwares.vercel.app/privacy.
Who it talks to
To answer, it makes requests to the site you ask it to scan, from a headless browser that signs up with a clearly labelled probe address at example.com and never solves a CAPTCHA. Those requests come from us and say so in their user-agent; they carry nothing of yours.
Who else touches it
- Vercel, which hosts it in the United States. Vercel keeps its own log of each request — its path and query string (so the name or URL in a check made over the web API), the requesting network address and the user-agent — for one day on our plan, and we read it only to fix faults. The arguments of an MCP tool call travel in the request body, which we do not log.
- Neon, which holds the counted page views and what is listed above, in the United States.
- GitHub, whose Actions runner executes the browser in a public repository, so the run, which names the URL scanned, is public too.
- Our model provider (Anthropic), which reads the scanned site's public terms and docs pages to answer two questions about them. Nothing beyond those public pages is sent, and they are not used to train their models under the terms we use.
How long we keep it
- Counted rows identify nobody, so they are kept without an expiry date, as the running tally of how the tool is used.
- Vercel's request log: one day.
- Scans are kept, and stay public, until the domain's owner asks us to remove them at https://agentwares.vercel.app/contact.
What we do not do
We do not sell data, set cookies, build profiles, or use anything you send to train models. We share nothing except with the processors above, and only to run the tool.
Your rights
Ask and you get a copy of anything we hold about you, a correction, or its deletion — whether or not the law where you live requires it of us. Ask at https://agentwares.vercel.app/contact. We aim to answer within 30 days.