Data Model

The Arrakis platform is built around seven core entities that capture the identity, behavior, risk posture, and operational context of every AI agent in an organization.

• Agent: The central entity representing a single autonomous AI agent, with identity metadata, governance state, risk score, and permissions tracking.
• Alert: A security finding raised when an agent exhibits risky or non-compliant behavior, with severity classification, remediation guidance, and compliance mappings.
• Tool: An API endpoint or service that an agent can invoke during execution, tracked with its provider and security posture.
• Integration: An external system connection that an agent interacts with, tracking connection status and data classification.
• Policy: A security rule that the Policy Engine evaluates against agent behavior, covering both built-in and custom policies.
• AgentTrace: An execution trace capturing a single run of an agent, with full observability into actions, tool calls, and timing.
• AgentType: A template defining common characteristics for a class of agents, enabling standardized governance across similar agents.

Agents generate alerts, produce traces, use tools, and connect to integrations. Policies are standalone entities evaluated against agents by the policy engine. Each agent belongs to one agent type.

Full data model documentation is available to Arrakis customers.