What is Combined Memory?
Combined memory is an architecture pattern that merges multiple types of memory systems to provide AI agents with both immediate working context and access to historical information. This approach typically integrates short-term memory (like conversation buffers) with long-term memory (like vector stores) and sometimes specialized memory types (like entity memory or summary memory) to create a comprehensive memory system that supports diverse retrieval needs.
The architecture allows agents to benefit from the strengths of different memory types simultaneously. For example, a combined memory system might use a buffer to maintain recent conversation turns for immediate context, a vector store to retrieve relevant historical information based on semantic similarity, and an entity store to track important facts about specific people or objects mentioned in conversations. When the agent needs context, it can query all these memory types and synthesize the results.
Implementing combined memory requires careful orchestration to prevent information conflicts, manage the relative priority of different memory sources, and ensure efficient retrieval across multiple storage backends. The system must decide how to blend information from different memory types when they provide overlapping or conflicting context, and how to stay within token limits when assembling context from multiple sources. Despite these challenges, combined memory architectures are increasingly common in sophisticated agent systems that require rich contextual awareness.