Vector Memory

Vector memory is agent memory stored as embedding vectors in a vector database, retrieved by semantic similarity. Each piece of memory (a turn, a fact, a chunk of a document) is encoded into an embedding with a chosen model and stored alongside metadata; recall queries are also embedded and the closest matches returned. It is a common backbone for both episodic-style memory (every turn embedded) and semantic-style memory (each extracted fact embedded). Limits: recall quality depends on embedding model and chunking, structured updates are awkward, and similarity does not equal relevance for some query types.