Privacy

The replay makes no model call. Data goes to the Meta Model API only during a live run that a visitor starts on the Live run page.

What a live run sends to the Meta Model API

• Screenshots of the allowlisted demo page being walked (a W3C Before and After Demonstration page or a builder-made page), 1024 by 768 pixels: for each Enter pressed on a button, the screenshots before Enter, after Enter and after Escape.
• With them, the focused element's role, accessible name and focus box (its position and size on the screenshot), and the fixed reviewer prompt.
• At most one text-only call for the packets, when the page has findings: the findings on that page (rule, key step, element role and name, focus boxes) and the fixed writer prompt.
• The task you type, if any, and up to five screenshots per key: with a task, the keyboard navigator asks the model for each key (at most 41 calls), sending the task text, the keys pressed so far and the latest screenshots of the page, but no focus box, role or name. Do not type personal information into the task.
• Nothing else about you: no address you type (only listed pages can be walked), no cookies, no other form input and no network address.

Every call uses the model muse-spark-1.3 on the Standard tier (not the Contributor-mode variant), reasoning effort minimal, and asks the API not to store the response for later retrieval (store: false; the documented default is to store it).

What Meta says about use and retention

From the Meta Model API Data Commitments (read 2026-10-30):

Standard Services: Meta does not use your Content from Standard Services to train Meta Models. Meta may process Content as necessary to detect, prevent, and respond to violations of the Model API AUP or the Terms of Service, and to develop and improve systems that protect the safety and security of Meta's products and services.

Data Retention. Meta will retain your Content as described in the Terms and Privacy Policy. You can view and delete your Content through your Account. Certain retention may apply as described in the Terms.

From the Meta Model API Terms of Service, section 4.4 Retention (last updated October 9, 2026):

Meta will retain your Content and Usage Data as needed: (i) to provide the Services under these Terms; (ii) to the extent necessary for compliance with applicable laws; (iii) in connection with the exercise of Meta's rights to use your Content and Usage Data under these Terms; (iv) when our systems flag data for a potential policy violation; and (v) for security, safety, abuse, and policy review and improvements.

We found no fixed retention period (a number of days) in Meta's developer documentation at dev.meta.ai/docs or its legal pages as read on 2026-10-30. Meta's Privacy Policy also applies.

What this server keeps

• Each live run (at most 200 in all): its screenshots, focus record, frozen verdicts, model answers and packets, and the typed task, shown in the replay viewer at the run's link.
• A credit ledger with, per model call, the time and run id of a cost reservation written before the call is sent, then the token counts and the cost once it is answered. No prompts, images or request headers.
• Start times per network address, in memory only, to limit how many runs one address can start; an address's times are dropped at the first run start more than an hour later, and all of them when the server restarts. Nothing is written about visitors, and there are no cookies or analytics.

Contact

Questions or requests: open an issue at https://github.com/javierdejesusda/nomouse/issues.