Have you considered throwing Gemini into the mix? Given a context window of 2M, i.e. 10x Claude's, it might have at least "mid-term" which might prove useful. For long-term memory you probably need a vector database system like you describe. We have all the ingredients in terms of infrastructure, algoritms and methods so it should be a question of implementation, which likely requires some time with the current state of things. However, once AI reaches escape velocity, i.e. it is smart enough to implement itself, you should be able to pass this post as a prompt and your desired system should pop into existence :D.
My ai-code-review has Gemini as an option, and it works very well. I haven’t found a good agentic tool that supports Gemini (yet) but I’ll keep looking.
Have you considered throwing Gemini into the mix? Given a context window of 2M, i.e. 10x Claude's, it might have at least "mid-term" which might prove useful. For long-term memory you probably need a vector database system like you describe. We have all the ingredients in terms of infrastructure, algoritms and methods so it should be a question of implementation, which likely requires some time with the current state of things. However, once AI reaches escape velocity, i.e. it is smart enough to implement itself, you should be able to pass this post as a prompt and your desired system should pop into existence :D.
My ai-code-review has Gemini as an option, and it works very well. I haven’t found a good agentic tool that supports Gemini (yet) but I’ll keep looking.