Definition
Semantic memory in AI agents refers to the long-term, persistent storage of generalized facts, concepts, and world knowledge, decoupled from specific temporal events. In RAG pipelines, this is implemented via vector databases that store high-dimensional embeddings, allowing agents to retrieve conceptually relevant information through similarity search rather than keyword matching.
The 'General Knowledge' (facts) of an agent, as opposed to Episodic Memory which tracks 'Sequence of Events' (chat history).
"An infinite library where books are shelved by the 'meaning' of their content rather than by title, author, or the date they were acquired."
- Vector Database(Storage Infrastructure)
- Episodic Memory(Temporal Counterpart)
- Embedding Model(Prerequisite)
- Knowledge Graph(Structured Alternative/Enhancement)
Conceptual Overview
Semantic memory in AI agents refers to the long-term, persistent storage of generalized facts, concepts, and world knowledge, decoupled from specific temporal events. In RAG pipelines, this is implemented via vector databases that store high-dimensional embeddings, allowing agents to retrieve conceptually relevant information through similarity search rather than keyword matching.
Disambiguation
The 'General Knowledge' (facts) of an agent, as opposed to Episodic Memory which tracks 'Sequence of Events' (chat history).
Visual Analog
An infinite library where books are shelved by the 'meaning' of their content rather than by title, author, or the date they were acquired.