N.KOJIRA.IO
Timeline
It's a multilayered approach and it is much more versatile than frontier model approaches.

So far I have 3 layers.

1. Pre-prompt with a pasted copy of my core context.
This is injected into my own models and works surprisingly well with both dumb and more capable models.

2. Knowledge attachment. This is the big one and the one I've built over 2 years. I have developed a hierarchy of context and I can load parts or all of it depending on the task. See below:

3. Persistent memory. An experimental feature, but a very useful one, and is shown below:

https://i.nostr.build/gJRpazDMPnfz17sKhUUNp7.jpg

https://i.nostr.build/qvlOiHlWkcph0PWHwebocr.jpg


https://i.nostr.build/5xswyJCg0exiHt3OmzEP87.jpg