Bug Report: New chat memory is very ineffective so far
`Steps to Reproduce`
Prompt: In another chat in this project you suggested .... Do you remember the specific example?
`Expected Result`
Model should search the current project, not entire chat history. Whatever scope it uses, it it does not find an answer in vectorized summaries it should search plain text. If it still has no answer it should simply say so.
`Actual Result`
The model searches the entire chat history, not just the current project. That's expected as I don't think project-specific scoping is supported. But the model considers a lot of chats and seems to drown in options - also no surprise in a lot of history. The result is a hallucination about text, an incongruent response that doesn't match the query, and other weirdness. It's just not ready for primetime.
`Environment`
Windows 10, Firefox, o3 and 4o models
devout marshBOT