This is cool. I used Jev to improve my custom memory system. Results: → 94% fewer tokens used → 2–3× faster memory retrieval This was just a quick test, but I believe it opens up so many new opportunities.

very cool

Another great idea for Jev utilization. well done.

How much more efficient is this than RAG + vector database?

Interesting idea, what is special about your "custom memory system"? Why not use Mem0 or Honcho?

i’ve tried mem0 and fabric but it’s kind of overkill for my use case sth like QMD might actually be useful, not sure if that would be better than this Jev system, i’d have to test it

How I use it ?

does retrieval stay stable as memory grows?

this is the part I’d want to measure next: does the 94% token reduction in memory retrieval actually translate into less repeated context in the LLM loop? I’ve seen a coding-agent session hit 99% repeated context, so I’m curious how much of that can actually be eliminated by a better memory layer.

This might be one of the best Jev use cases I’ve seen. Don’t make the big model read the whole memory. Let Jev decide what’s relevant first, then spend tokens only on the small slice that matters. 👀

Today I also did same. It is really fast and efficient.