4 Comments
User's avatar
Sydney Olson's avatar

This was an entertaining read! Definitely a step in the right direction towards AIs that act as agents without being re-prompted and trained each day. I often find myself copying and pasting the same CETO-framed responses each time I begin a new chat... would be nice to be able to delete that sticky note off my desktop.

David Heijl's avatar

That’s exactly the kind of thing these memory systems aim to replace, Sydney. I have to admit I’m a little curious to find out more about the CETO framed prompt you mentioned - what does that entail?

Suny Choudhary's avatar

This is such a neat way to think about it.

I'm Curious, how are you planning to handle memory conflicts when multiple agents write to the same context?

David Heijl's avatar

I set up a weighting system for which models are more trusted - you can configure it in Vist preferences. I wanted to cover my own use case: I enter plan mode, use Claude Opus for big new features for examples; but for simple execution of previous planned work, I’ll use sonnet or Gemini Flash even. And because Opus is more trusted, it will win out in case of direct conflicts. Memories also age and that will also affect the outcome.

We’ll see how all that works!