Rohit

boris has shared a thread with these details

I personally love the --teleport feature using which I can make connect the context in my browser claude instance, my local instance and in mobile as well.

the only article you'll ever need to you claude code like a pro, fr fr

a friend sent me via pinterest, and it's ai tweaked. the original image is different

it is officially said by Boris cherny on X itself, i just made it simpler and detailed to understand

can you run 15 instances in antigravity parallelly across mobile, web, locally ?
and share context b/w all these sessions ?

will post the repo which can be self hosted soon

not really, i am not claiming i invented it. i have just shared my learnings after building a failed system

hahaha, that's what differentiate AI from AGI.

I don't have a job, mate 🫂... and yes there is a way to timestamps in Vector DB which works well if you have implemented the overwriting the records correctly

thanks paul, for the classify_item funtion you can use an llm with predefined category list as guidance, but also write a function to make the system flexible enough to create new categories, i would recommend start with 5-7 fix categories and then let the system grow.

great repo, to learn from...honestly every open source repo has some or the other downside as of now whether its mem0 or any other company in that space...it's best to create your own memory context based on the need.

i have tried for even 1000+ messages and was able to get desired result...i will share another article(if i win over my procastination) on testing suite for long term memory

at 100K+ messages, we will use tiered summaries as
without tiered summaries, we would be searching through 100K embedded chunks which is expensive and slow but with summaries, you search maybe 20-30 high-level category summaries first, which is instant.