Unverified50% confidenceFactExact time
Summary Memory compresses old messages into a summary before they are deleted, and the summary is usually generated with a cheap small model to control cost; in LangChain this corresponds to SummarizationMiddleware().
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/18/2026
First Seen
Valid until: 1/16/2027
Sources
Related Entities
Related Claims
UnverifiedSummarizationMiddleware's basic logic is that when the number of messages or the token count exceeds a threshold, it calls a summarization model to compress earlier conversation into a summary, freeing up context window space.79% similarUnverifiedSummarization middleware is commonly used to compress long conversation histories or large documents in order to fit content within a model's limited context window.70% similarUnverified短期记忆的管理策略包括消息截断（只保留最近N轮对话）和摘要压缩（用LLM将早期对话压缩成摘要）70% similarUnverifiedSelective memory compression uses the decision model to determine whether a certain execution segment is 'closed'; if closed, it is collapsed into a semantic summary, otherwise the original log is retained.69% similarUnverified上下文压缩的核心思路是在保留关键信息的前提下对历史上下文进行精简或摘要，从而减少每次请求携带的Token量69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1060558API
curl https://kongchang.com/api/v1/knowledge/claims/1060558MCP
get_claim(id=1060558)