LLM knowledge base
Updated 2026-07-11
Questions this answers
- I have a pile of PDFs and articles and I want my AI to actually know them.
- How do I stop re-pasting the same source docs into every chat?
- Is there a simpler way to do research with AI than a full RAG setup?
- How do I build up knowledge over time instead of starting from scratch each time?
The Fix
Here's a research workflow you can start today, no fancy setup required. Instead of pasting sources into a chat over and over, you build a knowledge base the AI maintains for you.
It works like this. Drop your sources into a folder, something like a raw folder full of PDFs, articles, and notes. Then have the AI compile them into a wiki: a set of linked Markdown files with summaries, backlinks, and short concept articles. You view the whole thing in a tool like Obsidian, which is built for browsing linked Markdown notes.
The nice part is it compounds. When you ask a question, the AI answers against the whole wiki, and you file the answer back in as a new note. New sources get compiled in the same way. Every query makes the base a little richer, and you never hand-edit the wiki yourself. It's a lightweight alternative to setting up a full retrieval system (the technical name for that is RAG, retrieval-augmented generation).
When to Use It
Reach for this when you have a body of material you keep coming back to: research for a project, a competitive landscape, a topic you're going deep on. It's overkill for a one-time question you'll never revisit. The payoff shows up when the same sources get queried again and again.
In the Wild
Meet Cabinet: Paper Clip + KB. for quite some time I've been thinking how LLMs are missing the knowledge base - where I can dump CSVs, PDFs, and most important - inline web app. running on Claude Code with agents with heartbeats and jobs runcabinet.com
LLM Knowledge Bases Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating
Walks through compiling sources into a linked Markdown wiki and querying it in Obsidian without a full RAG stack.
Cabinet
runcabinet.com
A tool built around this compile-and-query-your-sources workflow.