Runtime Agents

Runtime agents are Adam-managed AI actors that can have governed model access, tools, approvals, work metadata, and trace evidence. They are distinct from service agents: a service agent establishes an Eden principal, while a runtime agent represents the AI workload and its operating policy.

Use a dedicated gateway key, explicit model routes, least-privilege tool access, and safe work metadata for each independently governed runtime.

Last updated: July 10, 2026