AI Agents Inventory and Attributes

Description — The AI or machine-learning model the agent is built on and driven by.

Benefit(s) — Exposes model dependency and concentration risk; the most defining technical relationship.

Source — Derived.

Notes — Links to the AI and Machine Learning Models Inventory. The model is not the agent.

Description — The specific version of the underlying model in use.

Benefit(s) — Ties behavior and evaluations to an exact model version and flags exposure when a version is deprecated.

Source — Derived.

Framework and Runtime

[Multi-Value]

Description — The agent framework, orchestration library, and runtime the agent is built and runs on.

Benefit(s) — Reveals technology dependencies and lock-in at the agent layer.

Source — Derived.

Examples — Agent orchestration framework; managed runtime environment

Notes — Links to the Software Technologies Inventory.

Description — The governing instruction set that defines the agent’s role, constraints, and behavior.

Benefit(s) — Makes the agent’s intended behavior inspectable and governable — the closest thing to its “training.”

Source — Manual.

Notes — Reference or store per data-sensitivity policy; the prompt itself may be sensitive.

Description — Whether and how the agent retains state, conversation history, or context across steps or sessions.

Benefit(s) — Surfaces a distinct governance and risk surface — persistent memory can be poisoned, leaked, or retained beyond policy.

Source — Manual.

Examples — Stateless; Session memory; Persistent vector store

Notes — Phrased durably — describes behavior, not a specific product.

Communication Protocols

[Multi-Value]

Description — The protocols the agent uses to reach tools, resources, and other agents.

Benefit(s) — Reveals how the agent connects and what governs those connections.

Source — Manual.

Examples — MCP; agent-to-agent; REST

Notes — MCP connections are realized as Integrations and, where the server is a first-class asset, link to the MCP Servers Inventory.

Tools and Functions Available

[Multi-Value]

Description — The tools, functions, and APIs the agent can invoke to act.

Benefit(s) — Defines the agent’s action surface — what it can actually do.

Source — Derived.

Examples — search_orders; issue_refund; send_email

Notes — Reached through the Integrations Inventory.

Description — The pre- and post-model guardrails configured for the agent — input filters, output checks, and policy enforcement.

Benefit(s) — Documents the runtime controls that bound the agent’s behavior.

Source — Manual.

Notes — The governing policies behind these live in Governance; enforcement in Autonomy and Behavioral Controls.

Description — How the agent’s runtime is isolated so it cannot act beyond defined resources.

Benefit(s) — Limits blast radius — an agent that executes tasks must be contained.

Source — Manual.

Examples — Containerized; execution whitelist