Breaking

Karpathy's 'LLM Wiki' Draws Attention via Obsidian-Claude Setup

June 29, 2026 at 23:15 EDT

AI Agents
Software Dev & Coding

An 'LLM Wiki' pattern published by Andrej Karpathy, formerly of OpenAI and Tesla, is again drawing attention among developers as a way to link the note-taking app Obsidian with Anthropic's Claude Desktop in about 15 minutes, turning personal notes into a queryable 'second brain.'

April 2026 · Andrej Karpathy

The "LLM Wiki": Turning Obsidian Notes Into a Self-Maintaining Second Brain

Pair Claude with a local Obsidian vault and the LLM "compiles" raw sources once into persistent, interlinked Markdown — so knowledge compounds instead of resetting on every query.

54×

Faster CLI search vs grep on a 4,663-file vault

5–15 min

Setup time from prompt to working vault

Vector DBs needed — plain local Markdown

Search speed on a 4,663-file vault

Proportional — taller column means faster.

1×

grep
(baseline)

54×

LLM Wiki
CLI search

Three actions that run the vault

INGEST

Consolidate raw sources into interlinked wiki Markdown

→

QUERY

Search and synthesize across the accumulated knowledge

→

LINT

Detect contradictions and run health checks

The mental model

Obsidian is the IDE · the LLM is the programmer · the vault is the codebase.

raw/ — immutable source-material layer

wiki/ — LLM-generated interlinked Markdown

CLAUDE.md — schema + rules for the vault

WHY IT WINS

Knowledge compounds — no context reset per query
Fully local: privacy, no vector DB, low maintenance
Eliminates manual linking; surfaces forgotten notes
Graph view + backlinks for visual navigation

WHAT TO WATCH

Tied to a local device with Obsidian open
Large vaults raise token use and slow down
Hallucinations / bad edits — use Git for rollback
MCP / Filesystem setup is an initial hurdle

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.