AI Glossary · Agentic AI
Memory (Agent Memory)
Persistent state an agent carries across turns or sessions — short-term (scratchpad), long-term (vector store), episodic (past runs).
Definition
What is Memory (Agent Memory)?
Memory (Agent Memory) is persistent state an agent carries across turns or sessions — short-term (scratchpad), long-term (vector store), episodic (past runs).
- Category
- Agentic AI
- Glossary set
- 11 related terms
- Audience
- Enterprise AI leaders
Why does Memory (Agent Memory) matter in enterprise AI?
Memory (Agent Memory) matters in agentic AI programs because it shapes how autonomous workflows plan, use tools, escalate exceptions, and produce auditable outcomes.
Related terms in Agentic AI
- Agentic AI
- AI systems that pursue goals autonomously by planning multi-step actions, calling tools and APIs, and revising their approach based on feedback — rather than only responding to a single prompt.
- AI Agent
- A software entity built around an LLM that perceives context, reasons over a task, invokes tools, and produces an outcome — often on behalf of a user or another agent.
- Autonomy Level
- A taxonomy (advisor → co-pilot → supervised agent → autonomous agent) describing how much human oversight a workflow requires.
- Digital Employee
- A packaged agent (with a role, tools and SLAs) that performs a bounded function — collections, scheduling, tier-1 support — measured like a human FTE.
- Function Calling / Tool Use
- A model capability that produces structured JSON to invoke external functions, APIs or databases — the mechanism behind most agentic workflows.
- MCP (Model Context Protocol)
- An open protocol for exposing tools, data and prompts to AI models in a standardized way, simplifying agent-to-system integration.