Almost every item in this list (except Karpathy's LLM Wiki) is based around vector embeddings.
Vector embeddings were super-hot a couple of years ago, but I don't think they have sticking power.
The moment you have an agentic tool calling loop the idea of doing a big fuzzy embedding search and hoping you get back relevant results loses attraction. You can give your agent ripgrep and let it figure out how to find the right results all on its own.
The biggest downside of embeddings is that it's very hard to set a threshold score below which you ignore things. If you ask a vector index for 10 results ordered by similarity you'll get 10 results - but results 3-10 might be total junk.
I always dismissed this category as more “markdown engineering” but this opened my eyes to some genuinely interesting things. The AI Memory space is more varied than I expected.
Almost every item in this list (except Karpathy's LLM Wiki) is based around vector embeddings.
Vector embeddings were super-hot a couple of years ago, but I don't think they have sticking power.
The moment you have an agentic tool calling loop the idea of doing a big fuzzy embedding search and hoping you get back relevant results loses attraction. You can give your agent ripgrep and let it figure out how to find the right results all on its own.
The biggest downside of embeddings is that it's very hard to set a threshold score below which you ignore things. If you ask a vector index for 10 results ordered by similarity you'll get 10 results - but results 3-10 might be total junk.
[flagged]
I always dismissed this category as more “markdown engineering” but this opened my eyes to some genuinely interesting things. The AI Memory space is more varied than I expected.
What about GBrain
[flagged]