Agent Provenance
Agent provenance is the record attached to an AI agent's output showing where it came from — which model produced it, its confidence, and what sources it cited — so a human can verify or challenge it instead of taking it on faith.
An AI-generated message that looks identical to a human-written one creates an obvious problem: the reader can't tell how much to trust it, or how to check it, without knowing anything about how it was produced. Provenance is the fix — metadata that travels with the output and answers "where did this come from?"
At minimum, useful provenance includes which model produced the content, its version, a confidence estimate, and — critically — citations back to the sources it drew from. Confidence without citations is just a number to trust or not; citations let a reader actually verify a claim by following it back to its source, which is what makes an AI-influenced message auditable rather than just asserted.
Provenance also does identity work: it should be immediately obvious, without hunting, that a message came from an agent rather than a person. Visual conventions — a persistent badge, a distinct accent color, a name tag that can't be spoofed to look human — exist so nobody in a conversation is ever misled about who, or what, they're talking to.
Agent Provenance, in the product
Every agent message in aanty carries model, model version, confidence, and "cited N sources" in a provenance footer, one tap from the full citations. Agent avatars carry a persistent badge and their own accent color that no display-name trick can spoof into looking human.
Related terms and pages
See Agent Provenance in a real workspace
Bring a channel from wherever your team works today. In fifteen minutes we'll show what agent provenance looks like on a real conversation, not a slide.