About

Karpathy LLM Wiki

An open-source Obsidian plugin that turns your notes into a self-organizing wiki — implementing Andrej Karpathy's LLM Wiki concept with local-first AI.

What is Karpathy LLM Wiki?

Karpathy LLM Wiki is a community plugin for Obsidian that implements Andrej Karpathy's LLM Wiki concept. You write notes in your vault as you normally would. The plugin reads those notes, extracts the people, concepts, and theories inside them — then generates interconnected wiki pages with bidirectional [[links]]. You can ask your wiki natural-language questions, and every answer links back to the source notes it came from.

The plugin is free, open source, local-first, and works with any Obsidian-supported LLM provider — including local models via Ollama or LM Studio.

The idea behind it

In April 2026, Andrej Karpathy published a gist sketching a workflow where a personal LLM reads your notes and writes a continuously-growing wiki on top of them — a system that turns raw text into structured, cross-linked knowledge automatically. Every new note gets folded into a living index without manual effort.

Karpathy LLM Wiki ships that workflow end-to-end. The goal is the same as Karpathy's: stop losing what you read. Make every paper, article, and thought recoverable — and let the connections surface themselves instead of relying on your future self to remember to write [[links]].

How it works

  1. Write your notes. Keep writing in Obsidian as you always do. Your original notes are never modified.
  2. Ingest. Run an ingest command — the plugin reads your selected notes and asks your LLM to extract entities, concepts, and theories. A source record is created in sources/ with frontmatter pointing back to your original file.
  3. Wiki generation. The plugin creates new wiki pages under entities/ and concepts/, with bidirectional [[links]] between them and the source records.
  4. Query. You can ask your wiki natural-language questions; every answer links back to the notes that informed it.
  5. Maintenance. New notes are auto-ingested; contradictions are flagged; reviewed pages stay untouched.

For the full architecture, see the GitHub Discussions.

Project principles

  • Local-first. Your notes never leave your machine unless you explicitly choose a cloud provider.
  • Provider-agnostic. Twelve supported LLM providers — including Ollama, LM Studio, and every major cloud API.
  • No telemetry, no backend, no account. The plugin has zero server infrastructure.
  • Reviewed pages are sacred. Once you mark a page as reviewed, the AI never overwrites it.
  • Open by default. The code is released under the Apache 2.0 license (with DCO); every line is on GitHub.

Who built it

Greener-Dalii is the project maintainer and primary author. The following contributors (listed in alphabetical order by GitHub handle) have made significant contributions:

  • green-dalii — project maintainer and primary author
  • DocTpoint — hub-retirement crystallization signal, knn baseline, peer-link quality analysis, cold-start A/B evaluation
  • dmarchevsky — code contributions and review feedback
  • FrancoTampieri — code contributions
  • Indexed-Apogrypha (Matthew Harper) — source-slug fingerprint, pre-ingest requirements gate, Anthropic fallback fixes, and issue reports
  • YounianC (xiaochuanlu) — ingest status bar and Query Wiki side panel

This list is maintained in the NOTICE file and is intended for attribution purposes under the Apache 2.0 license.

If you want to help, check the issue tracker or jump into Discussions.

License

Released under the Apache 2.0 license with Developer Certificate of Origin. See the LICENSE and NOTICE files on GitHub for details.

Useful links