Ai wiki

[PondelekB]

Hello i would like to ask if anyone knows which model is best for combing through .epub .txt and .docx or .odt files to act as a smart wiki for my books, i write many short novels and some long but most of them are connected so i would love to have an assitant that knows what happened for example in book 3 during the Epilogue of book B6 , i can work out the indexing and queries but would love to know if there is a lightweight model that will do the job even if it takes it some time.

My specs are
i7 8750k
Gtx 1060 MaxQ 6gb
16 gb ram

If anyone can answer thank u in advance

[SirThinksAl0t]

Suggestions:
LM Studio;
HuggingFace;

Both can show you hardware compatibility with your hardware, and you can search and filter models, on the website or app.

[RaresKeY]

You may want to look at these:

• SAG: https://github.com/Zleap-AI/SAG
  Closest to a full pipeline, with document parsing, event and entity extraction, relationships, vector search, and links back to source passages.

• MiniRAG: https://github.com/HKUDS/MiniRAG
  Probably the best fit for your 6 GB GPU. It is built around smaller local models and mixes normal retrieval with an entity graph.

• Graphiti: https://github.com/getzep/graphiti
  Useful for timelines and changing relationships across books, such as characters, items, events, and consequences.

I would start with MiniRAG, then borrow ideas from SAG.

books
-> Markdown
-> characters/items/events/consequences
-> linked maps or graph
-> RAG over structure and original passages

A LoRA could later handle extraction and categorization, but I would keep the books in the external index rather than trying to make the model memorize them.

All three are permissively licensed, so you can clone, fork, self-host, modify, and use them commercially as long as you keep the required notices.