开了自动记忆，发长文直接爆刷！

[funnys1r]

14:56:45
info
Document exceeded context limit (400 the input length exceeds the context length), attempting chunking...
Document exceeded context limit (400 the input length exceeds the context length), attempting chunking...
14:56:45
info
Split document into 1 chunks for embedding
Split document into 1 chunks for embedding
14:56:45
info
Document exceeded context limit (400 the input length exceeds the context length), attempting chunking...
Document exceeded context limit (400 the input length exceeds the context length), attempting chunking...
14:56:45
info
Split document into 1 chunks for embedding
Split document into 1 chunks for embedding
14:56:45
info
Document exceeded context limit (400 the input length exceeds the context length), attempting chunking...
Document exceeded context limit (400 the input length exceeds the context length), attempting chunking...

[AliceLJY]

感谢反馈！

从日志看，长文触发了 embedding chunking（文本超过 embedding 模型的 context limit 后自动分块），每个 chunk 都会打一条日志，所以看起来像"爆刷"。

这个问题在主仓库已有详细的 bug 分析和修复方向：
👉 CortexReach/memory-lancedb-pro#244

临时 workaround（任选一个）：

1. 关闭自动记忆捕获："autoCapture": false，改为手动调用 memory_store
2. 关闭自动记忆召回："autoRecall": false，改为手动调用 memory_recall
3. 关闭 chunking："chunking": false（长文不会分块，但超长文本的 embedding 质量会下降）

修复会在主仓库推进，主要方向是 query 截断 + 超时保护。