Just an Idea

[auf1r2]

Would be nice to have memory context switch. I.e. top level switcher that switches entire context of the memory.

Let's say you have several isolated projects or knowledge areas each of them should have fully autonomous memory not shared with other projects. I.e. my work project memory, my pet project memory, my personal research memory, etc.

Real story: Sometimes I use local MedGemma to consult it about my healt. I want it to remember all important things across conversations. But I don't want it to be shared with my hobby project or my work project. So as a user I need to have an option to switch memory context across all my separated projects.

[thecodacus]

waiting for this #5 to finishes before implementing this

[saattrupdan]

This would be great, but maybe not switching the entire context, since often there would be general memories that are useful to all agents.

Here's an attempt at a solution: what if we simply store the LLM used to create a memory in the memory's metadata (this might already be stored, not sure). Then if we mark providers as "private" in the config, then all memories created by LLMs from those providers would only be accessible to other private LLMs, but the private LLMs would have access to everything?