179 questions
No questions match those filters.
← All questions
AgentsEasy· Memory in Agents
What are the different types of memory in AI agents?
This is one of the questions in the full AI/ML interview bank. Pro unlocks all 1789 questions; Premium includes the same bank plus the highest daily Practice limit.
See plansFour labels get used, though in practice two of them collapse into one implementation:
- Short-term / working memory — the current conversation, living in the context window. Gone the moment the session ends.
- Long-term memory — anything persisted outside the context window, usually a vector database, that survives across sessions and restarts.
- Episodic memory — a specific flavor of long-term memory: records of particular past interactions, retrieved when something similar comes up again. “Last week you asked about RAG pipelines, want to continue?”
- Semantic memory — general knowledge the agent has accumulated over time, independent of any one conversation.
In practice, most agents that ship combine just two things: the current context window for short-term, and a vector store for long-term — with episodic and semantic often turning out to be different retrieval patterns against that same store rather than genuinely separate systems. The four-way taxonomy is useful for thinking clearly about what to store and how to retrieve it, but it’s not four different databases in a typical implementation.