Agent Registry

How it works

Agents register three ways. Most teams start with guided registration in the dashboard — add an owner, a framework, and the tools an agent can reach in a short form. Teams shipping through CI/CD register agents directly from a deploy step, using the TypeScript or Python SDK, so every agent enters the registry the moment it goes live rather than after someone remembers to add it.

Neither path catches everything on its own, so the registry also watches for shadow agents — activity from agents that were never registered. When Prefactor sees traffic from an unregistered agent, it surfaces it so the agent can be brought into the same inventory instead of running invisibly alongside the ones you already track.

What the registry tracks

Each entry records ownership and version — who built the agent and which release is running — alongside its framework (LangChain, CrewAI, and the other frameworks Prefactor integrates with natively or via OpenTelemetry), the environment it's deployed in (development, staging, production), and the tools and data it's permitted to reach. Deployment status moves the agent through its lifecycle to eventual retirement or archival, so a decommissioned agent stops showing up as live in dashboards and reports.

Version and environment are what make comparison possible elsewhere on the platform: live tracing and quality assessment both compare an agent's spans across versions and environments to catch drift, and that comparison only works because the registry already knows which version and environment produced each span.

Integration with your stack

Registration is designed to sit inside the deploy step you already have, not add a separate one. Call the registration API from CI/CD using the TypeScript or Python SDK, and an agent registers itself as part of the same pipeline that ships it — no manual step for teams that deploy agents frequently.

Because the registry underpins observability and enforcement alike, it's the first thing to wire up when adopting Prefactor: agent identity, framework, version, and environment are visible from the SDK integration itself, so registration usually needs no extra instrumentation beyond what tracing already requires.

From registry to everything else

The registry is the foundation for every other Prefactor capability. Live tracing and quality assessment compare spans by the version and environment the registry records. Cost attribution uses registry metadata to allocate spend to teams. Runtime policies reference the registry to determine which rules apply to which agents. Audit trails link every action back to the registered agent that performed it. Without the registry, none of the rest has an agent to attach to.