Unverified50% confidenceFactExact time
Context compaction is a class of techniques that distill existing conversation history into a more concise representation, typically by having the model generate a summary of earlier content to replace the original message list.
1
Sources
50%
Confidence
Long-term
Relevance
10/4/2026
First Seen
Sources
Related Entities
Related Claims
Unverified对话摘要记忆通过LLM将历史交互压缩为摘要，每当新消息加入时摘要会被迭代更新78% similarUnverified压缩功能更适合处理冗余日志，而非承载关键推理链的对话历史76% similarUnverifiedThere are three common strategies for context compression: truncation (retaining only the most recent messages), filtering (discarding irrelevant messages), and summarization (compressing older conversations into a single sentence).76% similarUnverifiedIn engineering practice, combining pruning, filtering, and summarization strategies is a common approach: retaining the original text for the most recent rounds of dialogue, generating summaries for earlier conversations, and directly filtering out small talk.76% 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75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/971441API
curl https://kongchang.com/api/v1/knowledge/claims/971441MCP
get_claim(id=971441)