A factual inventory of the agentic systems already in your environment.

Lens separates autonomous agents, agent-capable tools, model runtimes and supporting development systems. Its technical inventory can include frameworks, MCP servers, skills, models, API services, workflows, repositories and workloads without pretending that every framework import is an agent.

• Endpoint collectors inspect known configuration roots, processes and listeners.
• Repository and CI scans inspect manifests, imports and validated agent, MCP, A2A, OpenAPI and workflow descriptors.
• The Kubernetes controller observes workload metadata with read-only permissions and no Secret or pod-exec access.

Every finding is backed by sanitized evidence. Authoritative descriptors can confirm a system; otherwise Lens requires independent evidence families before raising confidence. Possible findings remain explicitly possible rather than becoming risk scores.

Coverage records what was checked, what was unavailable and where a collector reported partial results. That prevents an inventory count from being mistaken for complete organizational coverage.

Lens accepts inventory identifiers such as hostnames, repository names and sanitized endpoint hosts, while rejecting prompt bodies, source bodies, configuration bodies, credentials, environment values, secret values and full command arguments.

Local scans work without signup or network access and do not send telemetry. Organization-wide deployments can self-host Lens Hub and export Lens JSON, JSONL or CycloneDX 1.7.

npx barrikade-lens