Overview
Botscent helps a website determine whether an AI agent is browsing it and, when the available evidence is specific enough, identify the agent by name. It is an MIT-licensed tool with no dependencies, and its page component is about 5 KB when gzipped.
A verdict can report a named agent together with the reasons that produced the classification. If software appears to be an agent but does not match the registry, Botscent can still return an unnamed agent result.
How Detection Works
Botscent has server and page components that return the same verdict format. The server component runs in TypeScript or Python at the server or edge and examines the headers of a single request.
The page component observes browser-side evidence, which is useful for agents operating inside a person’s browser. Most sites use both components because declared or signed requests may be visible on the server, while browser-integrated agents may only expose evidence in the page.
Key Characteristics
- Recognizes named agents using documented signatures, user-agent values, and page-level markers.
- Includes a registry covering browser agents, user-directed fetchers, link previewers, crawlers, and other tools.
- Makes no outbound request from the server component.
- Makes no network request from the page component.
- Leaves logging, page changes, and access decisions to the website’s own code.
Detection Boundaries
A human verdict means only that Botscent observed no evidence of an agent; it does not prove that a person is present. The tool also does not claim to detect automation specifically engineered to appear human.
Botscent is a detection layer rather than a blocking system. Website operators decide how to use each verdict, such as recording agent activity or adapting their application behavior.