The local librarian — the next question for Guru

Research direction; results not yet established.

The next question is whether a dedicated local AI librarian can do more than preserve what I give it. Can it read new material, recognize a disagreement, revisit an earlier conclusion, and leave the knowledge in better condition?

My exploration of a local LLM and a custom inference engine was inspired by Charles Sullivan’s ai-research project ↗. His work prompted me to consider how Guru could extend into a persistent local librarian.

There is plenty to sketch and much less that has been proven. A system can accumulate an impressive library and still give poor answers. I want to measure whether additional reading improves its handling of unfamiliar questions, especially when sources disagree.

This is an experiment I intend to document as it develops, including approaches that turn out to add complexity without improving the result.