Agent Registry
An agent registry is the authoritative inventory of every AI agent in an organization: what each one is for, who owns it, what it may access, what it is approved to do, and which version is running in production. It answers "what exactly is running?" — the question most enterprises cannot answer today.
What an agent registry means
Agent counts in enterprises are growing faster than anyone can track with spreadsheets. The registry is the fix, and it does four jobs. Inventory: every agent — including ones built on different frameworks — appears once, with its owner, use case, data class and risk tier. Approval record: the registry links each agent to the review that authorized it, so "who allowed this agent" has an answer with a date and an approver attached. Deployment truth: the registry names exactly what is in production — the image fingerprint, the model versions, the tool grants — so drift between what was approved and what is running is detectable. Lifecycle: registration precedes deployment; promotion to production is an explicit act by a named approver; retirement is recorded rather than silent. Framework-agnostic registries matter because enterprises run agents on several stacks at once, and the governance question is always about the estate, not one framework's agents.
Frequently asked questions
Why does an enterprise need an agent registry?
Because "what exactly is running?" must have an answer before the first incident, audit or exam. The registry is the inventory that identity, policy and evidence all reference — without it, governance decisions have no authoritative list of agents to apply to.
What does an agent registry record?
Agent identity and owner, business purpose and annual value, data classification, risk tier, approved models and tools, the current production version with its approval record, and lifecycle status. The registry names what is running — not what was last documented.