Could you add a manual extraction switch? The multi-round extraction always inexplicably fails to trigger, far exceeding the set rounds, yet no new memories are extracted. Please optimize!!
Set the minimum number of entries to twice the trigger rounds. There is currently a bug where, if it is not an integer multiple of two, the first extraction of a new conversation will take twice the set number of rounds.
Example:
Correct Example: Trigger rounds 5, minimum entries 10
Effect: Triggered every 5 messages sent by the user (a total of 10 messages between the user and the assistant, and these ten messages are recorded. If set too low, it will not record completely.)
Bug Example: Trigger rounds 5, minimum entries 11
Effect: The first trigger occurs on the user’s 10th message. After that, it triggers every 5 messages sent by the user.
I already have fifty that don’t trigger QAQ
It may have already been triggered, possibly because the content of the conversation does not comply with the default summary prompt rules. The default “memory prompt” has content comparison and will not extract duplicate content.
It hasn’t been triggering recently, even after making changes according to your suggestions. It’s clear that it doesn’t meet the extraction and addition criteria. I disabled automatic extraction, changed it to one round, and it still doesn’t trigger. I don’t know what the issue is, but there is balance in the API.
…umm, I think I’ve found the problem; it’s due to the API specifications. The chat model uses Gemini, while the summary uses the default Silicon-based one, so it often doesn’t work.
There’s still an issue… the memory keeps being extracted repeatedly.
This topic was automatically closed 3 hours after the last reply. New replies are no longer allowed.