The AI Workspace is a centralized environment for defining and managing permissions for AI-enabled applications.
Each Workspace represents a real Application in your organization, whether an internal tool, customer-facing product, or shared service that uses AI capabilities. The Workspace acts as the authoritative boundary that defines what users and Agents are permitted to do.
Within an AI Workspace, you define:
- The Application context you are protecting.
- The Policies that govern how components in the agentic ecosystem can interact with that Application.
Security Model
The AI Workspace is built on a least-privilegeapproach. By default, users and Agents have no permissions.
Every capability, such as invoking an AI Action, accessing a protected Asset, or interacting with an external system, must be explicitly granted through a Policy with fine-grained conditions.
Policies define what is allowed, not what is blocked. Any access not explicitly granted by a Policy is denied by design. This model ensures that access is intentional, auditable, and aligned with business requirements.
Separation of Concerns
The following entities are defined and managed in their own dedicated Workspaces:
- Users.
- AI Agents.
- Tools, such as MCP servers.
- Data sources, such as RAG repositories.
The AI Workspace consumes these external definitions and applies Policy decisions to them without owning their lifecycle.
In the following sections, you will learn how to define and manage Policies that grant precise, least privileged access across your AI stack.