NVIDIACatProof is a member of NVIDIA Inception.Explore the program
catproof.
Get Started ↗
CATPROOF PRODUCT GUIDE

What is an AI agent
system of record?

A durable business record for an agent’s identity, accountable owner, purpose, lifecycle, and review decisions.

Looking for setup steps? Start with the technical documentation for hosted setup, connections and local development.

One agent can have several source identities.

A federated registry reads inventories from configured sources. Each source observation retains its platform, identifier, scope, and freshness. Operators link identities deliberately when they represent the same logical business agent; matching names alone are not proof.

Business ownership is a separate decision.

A platform owner may be a database role, maker account, or technical team. The accountable business owner is recorded independently. Provider refreshes preserve that ownership, purpose, lifecycle decision, and review notes.

Relationships make the record useful.

A relationship map connects the agent to its owner, source platforms, and related assets. Links should carry evidence and clear meaning. Available relationships depend on product support and the evidence the source provides.

Governance keeps decisions traceable.

Ownership, lifecycle, and review decisions belong with the business record and its history. Source observations describe what a provider reported; decisions describe what a person chose. Keep them distinct as the source changes.

Freshness is evidence with limits.

Last seen describes when a scoped observation was collected. It does not certify runtime health, safety, or compliance. Absence from one inventory read does not prove organization-wide deletion. Read failures should leave earlier evidence identifiable as stale.

Understand the product boundary.

CatProof provides registry records and decision history. A lifecycle marked suspended does not stop an agent in its source platform. The fictional demo makes no provider calls. Consult connector coverage for each adapter’s scope and validation status.

WORKED EXAMPLE · FICTIONAL DATA

One business record, two source identities

In the public demo, Ember represents a fictional invoice-triage agent. Two source identities are deliberately linked to one business record. The table shows the difference between a source observation and a business decision; it is not a customer deployment or a live provider recording.

FieldBefore simulated refreshAfter simulated refresh
Source identitiesdb-invoice-07 · sn-triage-12The same two identity links
Observed source nameInvoice triageInvoice exception triage v2
Business ownerJordan Lee, assigned in the demoJordan Lee, preserved
Decision historyOwner assignment recordedOwner decision plus source observation

Reproduce the fictional workflow: assign the owner, simulate a refresh, and inspect the history. This demonstrates the browser demo. It does not independently validate a production connector or backend persistence.

References and scope

Product explanations are first-party documentation. Provider references describe their interfaces; they do not establish a partnership, a completed tenant validation, or a CatProof performance result.

Follow Ember’s accountability story ↗

Explore connector coverage →