An LLM Wiki is a modern design pattern for Personal Knowledge (PKM). It treats a user's personal knowledge repository as a clean, interlinked, machine-readable "codebase" that growing AI agents can operate on over time.
This replaces fleeting AI chats and unstructured RAG databases with a persistent, compounding core of intelligence.
The LLM Wiki is a modern design pattern for personal knowledge management (PKM) popularized by Andrej Karpathy in early April 2026. It represents a paradigm shift from traditional Retrieval-Augmented Generation (RAG)—where an AI searches through unstructured files on the fly—to a compiled, persistent, and compounding knowledge base.
Karpathy's viral analogy
Karpathy summarized the ultimate relationship between the human, the AI, and the files with his famous analogy:
"Obsidian is the IDE; the LLM is the programmer; the wiki is the codebase."
He highlighted that traditional AI chat windows are "stateless" and fleeting, forcing you to reconstruct context every time you start a session. In contrast, an LLM Wiki compiles your unstructured raw inputs (diaries, PDFs, articles, research notes) into a clean, interlinked Markdown codebase ([slug]) that grows more intelligent and interconnected over time.
Farza's follow-up and Farzapedia
Developer Farza (founder of buildspace) created Farzapedia as the definitive real-world proof of concept for the LLM Wiki pattern:
-
The Input: Fed approximately 2,500 unstructured entries (comprising personal diary entries, Apple Notes, and raw iMessage histories) to an LLM agent.
-
The Compilation: The LLM agent parsed, categorized, and synthesized the raw materials, outputting ~400 beautifully structured, clean, and heavily interlinked wiki articles.
-
The Takeaway: Farzapedia demonstrated that an LLM can act as an objective, disciplined "gardener" of a human's mind, creating a private, local, and navigably explicit knowledge base that can be queried by any downstream AI.
Integrating the LLM Wiki in this wiki
This Wiki CLI repository is built directly on the principles of the LLM Wiki design pattern. It enforces:
-
Declarative Frontmatter: Structuring YAML-LD metadata to make pages machine-readable.
-
Procedural Automation: Using subcommands like Wiki Subcommand check to validate SHACL shapes and JSON Schema frontmatter, and Wiki Subcommand render to execute SPARQL queries dynamically, compiling the graph's intelligence back into static Markdown.
References
- Andrej Karpathy on X: LLM Knowledge Bases — April 2026 thread on raw ingest, LLM compilation into markdown wikis, Obsidian as IDE, and incremental Q&A without RAG.