LLM Wiki (nashsu/llm_wiki): What It Does and What It Still Needs From You
By Matt Cullerton, founder of Mavric and Apso
LLM Wiki is an open-source desktop app that reads your documents and has an LLM build and maintain a linked wiki from them, based on a pattern Andrej Karpathy published. A TikTok from @saf3synt describes its features accurately, then says it gives you organized knowledge "without manual effort." The project's own README says otherwise: it includes a human review queue, a periodic lint step and a stated "Human curates, LLM maintains" division of work.
What is LLM Wiki?
The repository is nashsu/llm_wiki, created on 2026-04-08 and maintained by nash_su (Yong Su). Its About text calls it "a cross-platform desktop application that turns your documents into an organized, interlinked knowledge base." It describes the difference from standard retrieval this way: "Instead of traditional RAG (retrieve-and-answer from scratch every time), the LLM incrementally builds and maintains a persistent wiki from your sources."
On 2026-10-07 the GitHub API reported 20,257 stars and 2,291 forks, more than the 19.0k shown in the video. The latest release then was v0.6.12, published 2026-09-28.
The @saf3synt account is a third-party curation account with the bio "Daily open source drops," not the developer.
How does the two-step ingest work?
The video says it uses "a two-step chain of thought process, analyzing content first, then generating wiki pages with source tracing built in." That matches the README.
The README describes "two sequential LLM calls." Step 1 produces a structured analysis of key entities, concepts and arguments, connections to existing wiki content, contradictions with existing knowledge, and recommendations for wiki structure. Step 2 generates a source summary, entity and concept pages, and updates index.md, log.md and overview.md.
Source tracing is a specific mechanism: "every generated wiki page includes a sources: [] field in YAML frontmatter, linking back to the raw source files that contributed to it."
Other listed features include a knowledge graph that ranks relevance on four factors (direct links, source overlap, Adamic-Adar and type affinity), Louvain community detection for finding clusters, KaTeX math and Mermaid diagrams, and a Deep Research mode.
Does LLM Wiki work without manual effort?
No, by the README's own description. Three parts of it involve a person:
- Section 12, "Review System (Async Human-in-the-Loop)," describes "an asynchronous review queue" where the "LLM flags items needing human judgment during ingest" and the "User handles reviews at their convenience."
- The Quick Start ends with "Check Review for items needing your attention" and "Run Lint periodically to maintain wiki health."
- The design principles name "Human curates, LLM maintains" as the fundamental role division.
Ingest is automated. Keeping the wiki accurate still takes your time.
How to install LLM Wiki
It is a desktop app, which the video does not mention. Pre-built binaries ship as .dmg for macOS on Apple Silicon, .msi for Windows, and .deb or .AppImage for Linux. Building from source needs Node.js 20+, Rust 1.88+ and protoc, then npm run tauri build. The codebase is mostly TypeScript with a Rust backend, which fits a Tauri app.
After installing, Quick Start step 2 says "Go to Settings → Configure your LLM provider (API key + model)." Supported providers are "OpenAI, Anthropic, Google, Ollama, Custom."
Is LLM Wiki free?
The app is free to download. Inference is billed to whichever provider key you enter, unless you run a local model through Ollama.
The LICENSE file contains the full GNU GPL v3 text with a prepended copyright line naming Yong Su for 2024-2026. That extra line is why GitHub's API reports the license as NOASSERTION instead of GPL-3.0. The project is GPLv3.
Where the idea came from
The README credits Andrej Karpathy: "The foundational methodology comes from Andrej Karpathy's llm-wiki.md, which describes the pattern of using LLMs to incrementally build and maintain a personal wiki." The gist was created on 2026-04-04, four days before the repository. It calls itself "an idea file, it is designed to be copy pasted to your own LLM Agent," and proposes a three-layer split of raw sources, the LLM-generated wiki and a schema. The video and its description do not mention Karpathy.
The README also claims "overall recall improved from 58.2% to 71.4% with vector search enabled," without a linked method or dataset.
Bottom line
LLM Wiki is a real, GPLv3 implementation of Karpathy's LLM wiki pattern, and its two-step ingest and source tracing work as the video says. Plan to supply your own API key and to spend time on its review queue and lint passes.
Sources
- nashsu/llm_wiki on GitHub - the repository
- GitHub API record for nashsu/llm_wiki - star count, creation date and license field
- README.md - ingest pipeline, review queue, install and providers
- LICENSE - GPLv3 text and copyright line
- Karpathy's llm-wiki.md gist - the original pattern
- LLM Wiki v0.6.12 release - current release at time of writing
- nashsu on GitHub - maintainer profile
- @saf3synt on TikTok - the posting account
Published September 12, 2026. Based on a TikTok by @saf3synt. Full video notes (markdown)