search

scanzy / magic3

This project creates some magic connecting a group of 3 entities: user, AI, bot.
MIT License
0 stars 0 forks source link

Reduce token usage #3

Open scanzy opened 1 year ago

scanzy commented 1 year ago

Use a different memory, to have longer conversations, without consuming a lot of openAI credits.

scanzy commented 1 year ago

Use different memory types depending on the current task type.

For example keep all the conversation within Q&A sessions, but summarize the previous output when switching to the next task.

scanzy commented 1 year ago

Try the ConversationSummaryBuffer in the first instance, then KnowledgeGraph