ProductFeaturesAgentsSecurityPricingIntegrationsCompareBook a demo
Feature · Agents

Agent Provenance

Every agent message is labeled — model, version, citations, confidence.

Sara Okafor triage-bot INV-2043 GitHub risk-agent Stripe // CONVERSATION GRAPH people · agents · entities
Overview

Inside Agent Provenance

You can always tell what produced a message and how sure it is. Agent output carries its model, version, grounding citations and a confidence signal, so an AI-authored message is never indistinguishable from a human-written one.

Labeled, always

Agent messages are visibly distinct from human ones — never disguised as a person typing.

Grounding citations

Claims an agent makes link back to the source they were grounded in.

Confidence, not certainty theatre

Agents surface how confident they are instead of stating everything with the same authority.

How it works

The detail

Every agent-authored message carries its model, version, grounding citations and a confidence signal, so it’s never mistaken for — or indistinguishable from — a human-written message. Provenance is part of the message, not a setting you have to go looking for.

Related

Keep exploring

Where this connects — in the product and in the rest of the catalog.

See Agent Provenance on your own workspace

Bring a real channel. In fifteen minutes we’ll show you agent provenance working on your team’s actual conversations, not a demo script.