What is Conversation Summary Memory?
Conversation Summary Memory is a memory implementation that maintains a running summary of the conversation history rather than storing the complete verbatim transcript. As new exchanges occur, the system uses an LLM to progressively update the summary to reflect the key points, decisions, and context from the conversation. This allows agents to maintain awareness of extended conversations without consuming excessive context window space.
The summarization process typically happens incrementally, with the system periodically taking the current summary and recent messages and asking an LLM to generate an updated summary that incorporates the new information. This approach ensures that the summary remains current and relevant while keeping its size bounded. The most recent messages are often kept in full fidelity alongside the summary to provide immediate context for the conversation.
Conversation Summary Memory trades perfect recall for efficiency and scalability. It enables agents to engage in very long conversations that would otherwise exceed context limits, but at the cost of potentially losing specific details that weren't deemed important enough to include in the summary. This memory type is particularly valuable for extended interactions like customer service sessions, tutoring conversations, or long-running collaborative tasks where the general context matters more than exact wording of every exchange.