Pinecone OR Redis❓ Best memory-systems.

[GoMightyAlgorythmGo]

Which is better pinecone or redis❓ (Quality; AutoGPT remembering what it previously did when on complex multiday project.)

(Ps: weaviate.io (!) & milvus.io also, i wish we could use all 4 and neural networks/statistics/autoGPT decide automatically, weaviate.io seems to have the best ideas. pinecone the best impression and wibe, redis the best reputation but seems not relly have those needed algorythms, it would be more good in combination, but it did suprisingly well so far....., and milvus.io sounds like a vector database, So not relly sounding interesting (altho again combination maybe)

[tlnet1981]

I use redis, but i dont think it ever pulls out anything of that memory, it keeps googling the same stuff, visiting same websites etc.

Anyone else?

[adriendarki]

I use redis too i think is better and free actually

[GoMightyAlgorythmGo]

for me it seems to never relly google the exact same stuff with redis. But it also seems to have memory issues for sure. I wonder if pinecone or https://weaviate.io/ might be better. Maybe pinecone gets better with time?

Edit: Pinecone definitely seems to have some positive effect on it - even in -debug mode i can tell it gives useful information quite frequently, would be amazing to combine it and improve or let different long-term GPT instances with long-term memory (like chatGPT chats don't disappear or you can resubmit prompt so there state of context is frozen in time) to look through it and get better and better memory's form pinecone.

[jiwoongim]

Hi, I am new to AutoGPT and going over the code, I notice that the permanent memory is not being used in the latest main branch (https://github.com/Significant-Gravitas/Auto-GPT/blob/preserve-permanent-memory-implementation/autogpt/chat.py#LL55C5-L55C17) , but contains in the old branch. Do you know what happened?

[Lushin415]

Which is better pinecone or redis❓ (Performance, AutoGPT remembering what it previously did when on complex multiday Project)

Have you managed to get AutoGPT to retrieve data from past operations from memory? Does he remember what he did?
Because with me he tries to solve the problem in a new way every time, without resorting to the previous analysis.

[GoMightyAlgorythmGo]

I noticed here that my AutoGPT uses the local memory of the computer, instead of Pinecone, I have already lost 70GB of hard disk space, and where everything goes I can’t figure out how to make it store information on pinecone. MEMORY_BACKEND=pinecone I have already specified such a parameter. I also encountered the following problem, autogpt does not want to read the files that I have prepared for it, for some reason it simply cannot find them, although I already gave it the full path of this form C:\user\name\downloads\autogtp\filename but it keeps returning error

Als chatGPT4 to set up everything. Maybe you need to give it a few tryes

[Lushin415]

I'm sorry, I understand very little in programming, so everything is quite difficult. But I figured out the memory problem.

1. I specified in the .env file MEMORY_BACKEND=pinecone (lowercase letters, not uppercase, it was PINECONE),
2. I added a wildcard index for OPEN AI. Then there was a message about using pinecone instead of local memory

[AntonioCiolino]

What do you mean by #2 above?

[GoMightyAlgorythmGo]

Which is better pinecone or redis❓ (Performance, AutoGPT remembering what it previously did when on complex multiday Project)

Have you managed to get AutoGPT to retrieve data from past operations from memory? Does he remember what he did? Because with me he tries to solve the problem in a new way every time, without resorting to the previous analysis.

Yes Definity have been able to do that. But not consistently enought. Pinecone seems to go on keywords or context that is similar to previous stuff therefore it should be combined in my opinion with other memory methods hardcoded and intelligent and adapting/lerning ones for optimal success.

[GoMightyAlgorythmGo]

I noticed here that my AutoGPT uses the local memory of the computer, instead of Pinecone, I have already lost 70GB of hard disk space, and where everything goes I can’t figure out how to make it store information on pinecone. MEMORY_BACKEND=pinecone I have already specified such a parameter. I also encountered the following problem, autogpt does not want to read the files that I have prepared for it, for some reason it simply cannot find them, although I already gave it the full path of this form C:\user\name\downloads\autogtp\filename but it keeps returning error

I suggest to use the command to tell autoGPT to use pinecone "python -m autogpt --use-memory pinecone"

[GoMightyAlgorythmGo]

I imagine only 1 memory system can be used

[aFastHero]

Unless two APIs are utilized. Rather than having the primary agent arbitrarily spawn agents under that same API, two main co-agents could be implemented, each with a dedicated API.