Most agent knowledge base tools today are "document dumps": throw everything into RAG and hope the agent picks the right info. If the agent gets confused or misinterprets sth? Too bad ¯\_(ツ)_/¯ you’re at the mercy of retrieval.
Socratic flips this: the expert should stay in control of the knowledge, not the vector index.
To do this, you collaborate with the Socratic agent to construct your knowledge base, like teaching a junior person how your system works. The result is a curated, explicit knowledge base you actually trust.
If you have a few minutes, I'm genuine wondering: is this a real problem for you? If so, does the solution sound useful?
I’m genuinely curious what others building agents think about the problem and direction. Any feedback is appreciated!
Hey everyone,
Most agent knowledge base tools today are "document dumps": throw everything into RAG and hope the agent picks the right info. If the agent gets confused or misinterprets sth? Too bad ¯\_(ツ)_/¯ you’re at the mercy of retrieval.
Socratic flips this: the expert should stay in control of the knowledge, not the vector index.
To do this, you collaborate with the Socratic agent to construct your knowledge base, like teaching a junior person how your system works. The result is a curated, explicit knowledge base you actually trust.
If you have a few minutes, I'm genuine wondering: is this a real problem for you? If so, does the solution sound useful?
I’m genuinely curious what others building agents think about the problem and direction. Any feedback is appreciated!
3-min demo: https://www.youtube.com/watch?v=R4YpbqQZlpU
Repo: https://github.com/kevins981/Socratic
Thank you!
Added support for any OpenAI-compatible LLMs, including locally hosted models and openrouter.