LLM Wiki

article Andrej Karpathy Apr 5, 2026

A pattern for building personal knowledge bases using LLMs. Instead of RAG-style retrieval from raw documents on every query, the LLM incrementally builds and maintains a persistent, interlinked wiki — a structured collection of markdown files that compounds over time. Three-layer architecture: raw sources (immutable), wiki (LLM-generated), schema (conventions). Three operations: ingest (process source into wiki pages), query (search and synthesize with optional filing back), lint (health-check for contradictions and gaps). The key insight is that wikis fail because maintenance burden exceeds value for humans, but LLMs eliminate that friction — they don’t forget cross-references or get bored updating pages. Traces the lineage to Vannevar Bush’s Memex (1945).

Backlinks

Agent Learning Loop The learning loop is the operational counterpart to the LLM wiki pattern (Karpathy). Where the wiki pattern applies ingest-compile-query to external knowledge, the learning loop applies the same cycle to the agent's own experience: experience is "ingested" into memory, "compiled" into skills, and "queried" via session search. →

Andrej Karpathy

Proposed the LLM Wiki Pattern — a pattern for LLM-maintained personal knowledge bases (LLM Wiki).

→

Andrej Karpathy AI researcher and educator. Co-founded OpenAI, led Tesla Autopilot's AI team. Known for deep learning research, educational content (YouTube lectures on neural networks), and practical contributions to AI tooling. →

Hermes Agent

Implements the LLM wiki pattern at the agent level — skills compound like wiki pages, memory persists like the raw layer
The 6 terminal backends directly implement the capsules concept — isolated environments that hibernate when idle
Skills system parallels CORAL's shared persistent memory — markdown files as accumulating knowledge
Context compression preserves structured summaries, similar to how the heartbeat mechanism in CORAL forces periodic reflection
Embodies the everything-is-text principle: all platforms, skills, memory, and config reduce to text

→

LLM Wiki Pattern A pattern proposed by Andrej Karpathy for using LLMs to build and maintain personal knowledge bases. Published as a GitHub gist (LLM Wiki). →

Knowledge Management

LLM Wiki — gist by Karpathy

→

Vannevar Bush and the Memex In his 1945 essay "As We May Think," Vannevar Bush proposed the Memex — a hypothetical device for storing, retrieving, and linking personal knowledge. The Memex was a desk-sized machine where a user could store all their books, records, and communications, and build associative trails between documents — personal paths through the material that could be shared with others. →