AI INVENTORY · AGENT REGISTRY
Every agent in your company — internal and external.
Agents in Kosmoy Action Capsules. Agents on Azure AI Foundry, AWS Bedrock and other platforms. One registry.
The Agent Registry covers what your company runs and what your company merely uses. Each entry has an owner, a runtime type (in-Capsule, external, scheduled), permissions, status and a link to the EU AI Act risk classification.
Inventory and risk classification cover everything. Runtime control — kill switch, JIT credentials, pre-flight authorisation — applies to agents running inside an Action Capsule.
What it does.
Internal + external
Same record shape, whether the agent runs in a Capsule or on Foundry / Bedrock.
Connectors to hyperscalers
Pull metadata and status from supported external platforms automatically.
Runtime type
In-Capsule, external (registered only), scheduled (cron / nightly), browser-based.
Permissions and scope
Which models, MCP servers and APIs each agent is allowed to reach.
Risk classification link
Each agent connects to its EU AI Act qualification, role and obligations.
Audit + lineage
Every change to scope, permissions or runtime type lands as an event.
Module questions, answered straight.
Does the Registry track agents on Foundry, Bedrock and other external platforms?
Yes. Internal agents (running in Kosmoy Action Capsules) and external agents (on Foundry, Bedrock, others) live in the same registry. Both get inventory and EU AI Act risk classification. Runtime control is reserved for what runs in a Capsule.
How does external-agent metadata stay in sync?
Connectors pull agent metadata from supported external platforms. Where the upstream platform exposes APIs, status and telemetry land in Kosmoy automatically. Where it doesn't, agents are entered manually and tagged accordingly.
What's the difference between an agent and an AI System?
An AI System integrates one or more models for a use case (chatbot, RAG, app). An agent is an autonomous component that calls tools, makes decisions and acts. A system can contain agents.
See the Agent Registry in production.
Walk through internal and external agent records, runtime status, and risk classification.