Episodic vs Semantic vs Procedural Memory in Agents
~15 min read
Long-term agent memory, mirroring human memory, splits into three distinct flavors: semantic (facts), episodic (past experiences), and procedural (learned how-to knowledge) — each stored and used differently.
Episodic vs Semantic vs Procedural Memory in Agents is a Pro topic
Sign in, then upgrade to Pro or Power to unlock this topic and the full AI Engineering curriculum.
Key points
- •Semantic memory stores facts/knowledge that are true independent of any specific event — typically a structured lookup
- •Episodic memory recalls specific past experiences or completed tasks — typically retrieved by similarity to the current situation
- •Procedural memory is internalized 'how-to' knowledge — for LLM agents, often refined prompts/instructions rather than retrievable records
- •Semantic memory answers 'what do I know'; episodic memory answers 'what have I seen before'; procedural memory answers 'how do I do this well'
- •This distinction directly informs storage architecture — database for semantic, vector store for episodic, versioned instructions for procedural