Agent Memory
conceptPersistent storage that lets agents recall past interactions, learn from experience, and build context over time.
Agent memory is what separates a stateful agent from a stateless chatbot. Memory systems store past interactions, learned facts, and agent reflections so they can be retrieved and used in future reasoning.
Common memory architectures include: short-term (conversation buffer), long-term (vector database for semantic retrieval), episodic (specific past events), and procedural (learned workflows). Many systems combine multiple types.
In SUBCORP, agents store memories in PostgreSQL with vector embeddings for semantic search. Memories are tagged by agent, type (observation, reflection, initiative), and importance. A memory archaeology system periodically surfaces old memories for re-evaluation, preventing knowledge from going stale.