Vannevar Bush & the Memex
Overview
Vannevar Bush was the originator of the Memex concept, introduced in 1945. The Memex was a vision of a personal, curated knowledge store in which documents were connected by *associative trails* — links between items that reflected how a human mind naturally moves from one idea to another. In Bush's model, the links between documents were considered as valuable as the documents themselves.
Source llm-wiki.en.md
The Core Idea
- A personal knowledge base, privately owned and actively curated by its user.
- Navigation driven by associative trails rather than hierarchical catalogues.
- The structure of connections was treated as a first-class artifact — not an afterthought.
The Unsolved Problem
Bush's vision had one critical gap: who does the maintenance? Building and continuously updating associative trails across a growing body of documents requires enormous, sustained effort from a human curator. Bush had no answer to this.
Connection to the LLM Wiki
Andrej Karpathy explicitly framed the LLM Wiki pattern as spiritually close to the Memex. The LLM solves precisely the problem Bush could never resolve:
*"It never gets bored, never forgets to update a cross-reference, and can touch a dozen pages in a single pass."* llm-wiki.en.md
In the LLM Wiki pattern, the large language model acts as the tireless curator — writing cross-references, surfacing contradictions, and keeping the associative structure consistent — while the human focuses on curating sources and asking questions. Tools like Obsidian make the resulting graph of wikilinks visually navigable, echoing Bush's vision of traversable associative trails.
Relationship to Other Concepts
| Memex concept | LLM Wiki equivalent |
|---|---|
| Personal knowledge store | The wiki directory of Markdown pages |
| Associative trails | wikilinks cross-references between pages |
| Human curator | The LLM agent (via Ingest / Query / Lint Workflow) |
| Private & owned | Git repo of Markdown files |
Unlike RAG, which re-derives answers from raw chunks at query time, the LLM Wiki compiles a persistent, interlinked structure — far closer to Bush's original vision of knowledge that *accumulates* rather than knowledge that is *retrieved on demand*. See LLM Wiki vs. RAG for a detailed comparison.
Managed services such as Cairni extend this pattern further, adding team workspaces, version history, and public publishing on top of the same compiled-wiki foundation. llm-wiki.en.md