graphkasten
Graphkasten
Blog
About

Search

Search IconIcon to open search

LLM Knowledge Bases

Last updatedUpdated: by Jakub Žovák · 2 min read

Properties
tags personalpersonal/pkm
created 04.04.2026, 19:00
modified 04.04.2026, 20:02
published Empty
sources Karpathy tweet 1 · Karpathy tweet 2 · llm-wiki.md gist
topics LLM Knowledge Bases, Organization Systems
authors Jakub
ai-assisted Yes

# Core Idea

  • Most RAG systems re-derive knowledge from scratch on every query — no accumulation
  • Instead: LLM incrementally builds and maintains a persistent wiki between you and raw sources
    • When a new source is added, LLM reads it, integrates it into existing wiki, updates cross-references, flags contradictions
    • Knowledge is compiled once and kept current, not re-derived on every query
  • The wiki is a persistent, compounding artifact — richer with every source and every question asked
  • You never write the wiki yourself — LLM writes and maintains all of it
    • Obsidian as the IDE; LLM as the programmer; wiki as the codebase

# Architecture

  • Raw sources — immutable source documents (articles, papers, images); LLM reads but never modifies
  • The wiki — LLM-generated markdown files; LLM owns this layer entirely
  • The schema — config doc (e.g. CLAUDE.md) telling the LLM how the wiki is structured and what workflows to follow

# Operations

  • Ingest — LLM processes a new source: summarizes, updates entity/concept pages, appends to log
  • Query — LLM searches wiki pages, synthesizes answer; good answers are filed back as new wiki pages
  • Lint — periodic health check: contradictions, orphan pages, stale claims, missing cross-references

# Indexing

  • index.md — content-oriented catalog; LLM reads this first when answering queries
  • log.md — append-only chronological record of ingests, queries, lint passes

# Use Cases

  • Personal tracking (goals, health, self-improvement)
  • Research deep-dives over weeks/months
  • Reading a book chapter-by-chapter
  • Business/team internal wikis fed by Slack, meetings, project docs

# Why It Works

  • Humans abandon wikis because maintenance burden grows faster than value
  • LLMs don’t get bored, don’t forget to update cross-references, can touch 15 files in one pass
  • Related in spirit to Vannevar Bush’s Memex (1945) — private, actively curated, associative trails

Backlinks