Try it free: dorothy.panaxon.ai. Upload a document and ask it anything.
For the past several months I've been building M.N.E.M.O. (Multidimensional Neural Elastic Memory Organization), a memory layer for AI that doesn't rely on vector search.
The problem
Similarity search always returns something, even when the answer isn't in your documents. The model then answers confidently from the wrong context, and you can't tell.
What I did differently
M.N.E.M.O. stores what you give it in a structured tree and looks up answers by exact path instead of by similarity. Two things come out of that:
- Every answer shows the path it came from, so you can check it.
- When the answer isn't there, it says so instead of guessing.
It doesn't remove every way an LLM can go wrong. What it removes is being handed the wrong context in the first place.
Where it's still weak
It depends on the model choosing the right path, and I haven't benchmarked that at large scale yet.
Try it and tell me where it breaks
A free account keeps your memory between sessions and lets you ask questions of knowledge other people have shared. Upload a document, ask something hard, then ask something that isn't in it, and see what happens.
If it gets something wrong, I'd really like to know. Leave a comment here.
Architecture and benchmarks for the curious: panaxon.ai
