Agent Memory

Agent memory is persistent or retrievable context used beyond a single model call. It can include documents, embeddings, graph relationships, task state, user preferences, and audit-safe history.

Memory is not the same as RAG. RAG is one use of retrieval; memory is the broader layer that manages what context can survive, be searched, and be reused.

Memory Responsibilities

• Retrieve relevant context for a task.
• Preserve provenance and visibility rules.
• Combine vector search with graph or metadata filters.
• Prevent untrusted retrieved text from overriding system policy.
• Support evaluation by making retrieval behavior inspectable.

Demo Mapping

The demo memory configuration is memory.yaml. It models:

• document chunking
• embeddings
• vector search
• knowledge graph expansion
• visibility filters

Related Pages

• RAG vs Memory
• Memory topic
• Operations and Evaluation