aichannel

Who’s working on agent memory layer specifically? Seems like a super fun problem to work on

If you’re working on agent’s memory specifically, come say hi!

designer alchemy | everything is experimental |
a decade+ of consumer products | farcaster godview: /intelligent | self exploration: /tanishq

i dabble with ai using open data from crypto https://alexpaden.tech @unbias

Just getting some reading in for memory management in agents — I think it could develop into an interesting space in itself

What about you!

I think a bit on providing general memory to many agents but not the individual agent memory systems 

Wondering if you’ve found anything novel compared to I.e similarity search

I stumbled-upon Zep, which in their own words is “A Temporal Knowledge Graph Architecture for Agent Memory”

They use something called Graphiti which is a graph based db structure

Faster than MemGPT and evolves beyond the current evaluation heuristics 

https://arxiv.org/abs/2501.13956

https://github.com/getzep/zep

https://github.com/getzep/graphiti

It’s tailor made for agent use case so has features that support that directly 

Temporal awareness is probably one such novel addition — adding time decays, the ability to focus on what context is still prevalent 

It’s also categorizing input into episodic, eventful and semantic memory

Is there anything meaningful to this beyond an I.e neo4j graph db? I’ll have to read when not mobile