CATPROOF PRODUCT GUIDE · UPDATED OCTOBER 27, 2026
What is an AI agent
system of record?
An AI agent system of record is an authoritative place for an organization’s agent identity, accountable owner, purpose and lifecycle decisions. A federated registry connects those records to observations from multiple source systems, keeping each observation attributable to its source.
What belongs in the canonical record?
In CatProof, the canonical record contains the agent’s name, purpose, business owner, local lifecycle, review state and notes. Source refreshes do not overwrite these decisions. Each linked observation retains its source platform, source ID, source owner, reported status and last-seen evidence.
How does federation work?
A connector reads an authorized scope and normalizes the observations. An operator deliberately links identities that represent the same logical agent. CatProof does not merge agents merely because their names match. A peer-registry connection reads a compatible canonical catalog through a separate, attributed connection.
For example, the fictional Invoice triage scenario in the interactive demo links a Snowflake identity and an A2A card to one record. That example is not customer evidence or a claim that those platforms synchronize automatically.
Why separate business owners from source owners?
A source owner may be a database role, a maker account or a platform team. That is useful evidence, but it may not identify the person accountable for the agent’s business use. CatProof keeps the source owner visible and stores the business owner as a separate local decision.
What does last seen tell you?
Last-seen evidence describes when a source observation was collected. It does not certify that an agent is serving traffic, safe or compliant. A failed or paused refresh retains previous evidence as stale. Absence from one successful scoped read does not prove organization-wide deletion.
Which platforms can I connect?
The connector catalog lists adapter scope, authentication and validation limits. Examples include Snowflake Cortex Agents, Microsoft Copilot Studio, AWS Bedrock Agents, Google Agent Registry and supplied A2A cards. MCP servers and directory identities remain related assets.
What does CatProof do today?
CatProof provides scoped read-only discovery, manual records, explicit identity linking, business ownership, lifecycle and review fields, history and export. Personal accounts have separate registries with owner, editor and viewer roles. Invitations are single-use codes shared by owners; delivery is manual. Protected connector credentials are provisioned per registry during early access.
CatProof does not invoke agents, enforce runtime permissions or provide compliance certification. Local lifecycle changes do not stop vendor agents. Email verification, emailed password recovery, SSO, email delivery and billing are not connected. Current snapshot persistence serializes requests and is capped at 25 MB per registry; it is an early implementation, not an unlimited enterprise deployment.