Vector memory
Also known as: semantic memory
Vector memory stores an agent’s memories as embeddings and retrieves them by semantic similarity.
- “What did we decide about auth last week?” → nearest vectors.
- Retrieves by meaning, not exact wording.
- The same machinery as RAG, pointed at the agent’s own history.
Why it beats keyword search: the agent phrases questions differently every time. Semantic retrieval is robust to that — the same underlying fact surfaces under many phrasings.