Instrument 04agentlens

The tool is the registry.

agentlens inventories the AI agents running in an environment and works out which of them nobody is watching. It is built for the case where there is no registry to check against, because that is the usual case.

Writeup

Two contexts, from the first day.

One is a single read-only diligence snapshot: the M&A shape, with the access a buyer actually gets and one exported artifact at the end of it. The other is a continuous series of snapshots with a diff between them.

Neither assumes a governance baseline exists. No AI gateway, no CMDB, no OpenTelemetry. agentlens builds the baseline rather than diffing against one, which is why the same tool serves both.

Monitoring status is derived, not looked up.

Whether an agent is monitored is inferred from observable properties of the agent itself: the path its traffic egresses by, whether its credential is attributable to a person or a service, where it was deployed from, who owns the compute it runs on, and whether independent sources corroborate each other about it.

None of that is a lookup, because there is nothing to look it up in. That is the premise the whole design rests on.

RegisterPublic surface

agentlens

Repository
Private, no public artifact
Modes
Single diligence snapshot · continuous series with diff
Collectors
13 across 7 providers · AWS · GCP · Azure · Keycloak · CrowdStrike · AWS SSM · operator attestation
Signals
Egress path · credential attributability · deployment provenance · compute ownership · cross-source corroboration