AI Agents

What is Memory in AI Agents?

Memory in AI agents refers to the ability to retain and recall information across interactions. Without memory, every conversation starts fresh. With memory, agents remember user preferences, past decisions, and accumulated context.

Why It Matters

Imagine an assistant that forgets your name every message, or a support agent that can't recall the ticket history. Memory enables continuity, personalization, and the accumulation of context that makes agents genuinely useful over time.

Types of Memory

Short-term (Working) Memory

Conversation history within current session. Limited by context window. Typically the full chat transcript.

Long-term (Semantic) Memory

Persistent facts and knowledge stored in vector DBs or knowledge graphs. "User prefers concise responses."

Episodic Memory

Records of specific past interactions. "On Jan 5, we discussed the Q4 report and user flagged Section 3 as problematic."

Implementation Approaches

  • • Full history — Include all messages (hits context limits)
  • • Summarization — Compress old context into summaries
  • • RAG retrieval — Store in vector DB, retrieve relevant
  • • Entity extraction — Extract and store key facts

Related Terms

← Back to Glossary