Memory (Agent) — AI Glossary

Reference details

Related terms

• Agentic Memory
• Episodic Memory
• Semantic Memory
• RAG (Retrieval-Augmented Generation)

Example: Memory is a retrieval problem wearing a different name

A support agent runs 40 sessions with one customer at roughly 12 turns each — about 480 stored entries. The next call retrieves the top 5 by similarity, so 475 are invisible, and similarity is not the ordering you want: the customer changed address in session 38, but session 3's address is textually just as similar and may rank above it. Working memory needs recency and precedence, not nearest-neighbour. This is why production memory splits into an episodic store of what happened and a small structured store of extracted current facts, where the address is a field that gets overwritten rather than a turn that gets re-ranked.

Commonly confused with

Three different things get called memory. The context window is what fits in one request and vanishes after it. Memory is state that persists between requests. RAG retrieves from a document corpus rather than from interaction history. A system can have a huge context window and no memory at all — that is the default, since the chat completions endpoint is stateless.

When to use it

Reach for it when:

• Assistants that must carry preferences or task progress across sessions
• Long-running agents, where the transcript would otherwise exceed any context window
• Where the same user returns and re-explaining is the main source of friction

Reach for something else when:

• Single-shot or stateless tasks — memory adds retrieval failure modes for no gain
• Without a forgetting and correction policy: stale facts are recalled as confidently as current ones
• For anything that belongs in a database — an order status is a lookup, not a memory

Continue your research

Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Memory (Agent) — AI Glossary.

• Browse models
• Review providers
• Inspect benchmarks
• Explore workflows
• Learn AI concepts
• Read decision guides
• Review trust and methodology
• Start a decision path