Agentic Assistant components and concepts

Learn about the key components and concepts, such as agents and actions in the Cortex Agentic Assistant

The Cortex Agentic Assistant uses the following components and concepts:

Name	Description
Actions	Actions wrap diverse capabilities (such as playbooks, scripts, and commands) to make them accessible and executable by an agent. You can use out-of-the-box system actions or register new actions.
Agent	An agent is a virtual persona that creates and executes domain-specific plans, at your request, to assist in your day-to-day SOC operations. An agent has roles and permissions that provide guardrails. Each agent is assigned a collection of actions that it can use as part of plans. The agent chooses the most relevant actions to fulfill a user's request. Agents process user requests, create plans, and orchestrate actions based on their goals and permissions (RBAC and SBAC). You can use the following types of agents: System agents that are provided for specific use cases. Custom agents that users have created. Some agents provide relevant chat conversation starters under the chat prompt. For examples of conversation starters, see Agentic Assistant use cases. Note Agents are bound by the same rules and robust permissions as a human user. In addition, you can mark actions that make real-world changes in production systems as sensitive, requiring a quick manual review and confirmation, ensuring peace of mind before critical system changes are made.
Plan	A sequence of actions that run in parallel or sequentially to satisfy a user request. The agent dynamically chooses relevant actions to resolve the prompt.
Conversation	A sequence of user requests that maintains context across interactions.
Knowledge	In the Knowledge Center, you can ground AI agents in your organization's business-specific source of truth by managing internal knowledge sources. You can also leverage built-in Cortex (system) knowledge to provide agents with deep product expertise and technical platform logic. This enables agents to act as context-aware extensions of your team that understand both your internal workflows and the underlying platform.
Request	A user request from the agent with an end goal, triggering a plan.