A RAG idea that stores knowledge as small trainable units
proposes storing knowledge as small concept-level units instead of plain document chunks. Each unit would hold facts, metadata such as source and , links to related concepts, and a small trainable part that captures how the concept should behave in context.
Standard RAG usually finds similar text chunks and sends them to a . would instead find only the relevant knowledge units, activate them, and use them to build the answer.
The idea overlaps with because it uses connections between pieces of knowledge, but it adds a LoRA-like trainable part for each concept. This is an early discussion idea, not a tested paper or product, so its effect on accuracy, token use, and cost is still unproven.
Key points
- Standard RAG often retrieves text chunks and feeds them into a model.
- would store each concept with facts, metadata, relationships, and a small trainable part.
- The proposed flow is to retrieve and activate relevant frames instead of sending raw chunks.
- The idea is related to but adds a LoRA-like learning per concept.
- It is still speculative, so cost and accuracy benefits have not been proven.