Unverified50% confidenceTradeoffExact time
多Agent并发且全程记录交互历史的工作方式会导致Token消耗大幅增长，因为每次请求都需携带完整历史上下文
1
Sources
50%
Confidence
Long-term
Relevance
7/20/2026
First Seen
Sources
Related Claims
Unverified长对话场景下上下文窗口会携带全部历史消息，Token消耗呈线性增长72% similarUnverified量化新分词器影响的最佳实践是从生产日志中随机抽取100-500条覆盖不同业务场景的请求，分别用新旧分词器计算Token数量进行对比62% similarUnverifiedThere are multiple strategies for updating long-term memory: Append preserves the complete history but storage costs grow linearly; Overwrite saves space but loses the trajectory of changes; Incremental Summary balances cost and information fidelity.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/572636API
curl https://kongchang.com/api/v1/knowledge/claims/572636MCP
get_claim(id=572636)