Terms of use

What AgentGlass does

AgentGlass measures whether AI browser agents can complete real tasks on a website, and publishes the method used to compute every score at /methodology so results can be recomputed independently.

Reports describe what a specific set of agents did against a specific URL at a specific time. They are evidence, not a guarantee: agent behaviour changes as models change, and a score is a measurement rather than a promise about future runs.

Requesting a scan

By submitting a URL for scanning, you confirm that:

• you own that site or are authorised to have it tested, and
• you accept that scanning drives a real browser against real pages, which produces ordinary traffic in your logs.

Scans are read-only by design. They do not attempt to create accounts, complete purchases, submit payment details, or modify anything on the site being measured. Don't submit a URL you do not control.

Reports and sharing

A report produced for you is yours to share. Some reports have a public link, and anything published at such a link is public — treat it that way before sharing one.

Acceptable use

• Don't use the site to scan, probe, or stress-test property you don't control.
• Don't attempt to disrupt the service or access parts of it you weren't given access to.
• Don't represent AgentGlass output as a security audit or a compliance certification. It is neither.

The usual, said briefly

The site and its reports are provided as-is, without warranty. Decisions taken on the basis of a report are yours. AgentGlass is not liable for indirect or consequential loss arising from use of the site.

Changes. These terms may change; the date below moves when they do.
Questions. hello@agentglass.dev — a real inbox, read by a person.