Semantic Memory
Semantic memory is memory of general facts independent of when or how they were learned. From cognitive science (Tulving, 1972), it is contrasted with episodic memory (specific events) and procedural memory (how-to skills). In LLM agents, semantic memory is the structured fact store: "user prefers metric units," "user's company has 47 employees," "user's primary language is Spanish." Implementations include extracted-and-deduplicated memories (mem0's stored memories), labeled persistent blocks (Letta's `human` and `persona`), structured records in a database, or knowledge graphs. It is distinct from episodic memory by stripping out the time/event context — semantic memory is the synthesized fact, not the conversation it came from.
Example
An assistant agent receives "I always run my calls in metric — switch the units in my workout summary." A semantic-memory layer stores `unit_preference: metric` as an atomic fact, scoped to the user. On every subsequent turn that involves units, the agent retrieves that fact and applies it — without re-asking and without scrolling back through episodic memory to find the original conversation.
Put this into practice
Build polished, copy-ready prompts in under 60 seconds with SurePrompts.
Try SurePrompts