fbpx

Hermes LLM Wiki: How to Set Up Karpathy’s LLM Wiki in Hermes Agent

WANT TO BOOST YOUR SEO TRAFFIC, RANK #1 & Get More CUSTOMERS?

Get free, instant access to our SEO video course, 120 SEO Tips, ChatGPT SEO Course, 999+ make money online ideas and get a 30 minute SEO consultation!

Just Click The Button Below To Get FREE, Instant Access!

The Hermes LLM wiki turns Hermes Agent into a research assistant that builds and maintains its own knowledge base: a folder of interlinked markdown pages that gets better every time you add a source. It comes as a bundled skill, llm-wiki, based on Andrej Karpathy’s LLM Wiki pattern, and it’s installed by default. In the video below I show the three layers, the three core operations and the setup. Here’s the full practical guide, from your first wiki to keeping it healthy.

Key takeaways

  • The llm-wiki skill ships with Hermes (research/llm-wiki, v2.1.0, MIT).
  • Your wiki is just markdown in ~/wiki (or WIKI_PATH) — no database.
  • Three operations: ingest, query and lint.
  • It opens as an Obsidian vault. Claim a free strategy session to plan your setup.

What the Hermes LLM wiki is

The idea comes from Karpathy’s LLM Wiki gist, which he calls an “idea file”: instead of asking an AI to rediscover your documents every time you ask a question, you let it build a persistent wiki that compounds. Hermes turns that idea into a bundled skill. The docs list it under research as llm-wiki, version 2.1.0, MIT-licensed, for Linux, macOS and Windows, with the obsidian and arxiv skills as related skills.

The SKILL.md sets out a clear split. You curate the sources and direct the analysis. The agent summarises, cross-references, files pages and keeps everything consistent. Unlike retrieval-based search, the knowledge is compiled once and kept current, so answers get better as the wiki grows. I covered the bigger picture earlier in Hermes AI LLM wiki integration; this page is the hands-on setup.

How to start your first wiki

Every installed skill in Hermes becomes a slash command, so you can run /llm-wiki. You can also just ask: “start a wiki about AI agents”, “ingest this article” or “lint my wiki”. Hermes picks the skill up when you create a wiki, add a source, ask a question the wiki can answer, run a health check or mention the wiki.

Choose where it lives with the WIKI_PATH environment variable, for example in ~/.hermes/.env. If you don’t set it, the default is ~/wiki.

# ~/.hermes/.env
WIKI_PATH=~/wiki

On the first run the agent works out the path, creates the folders, asks what the wiki is about, writes SCHEMA.md, index.md and log.md, and suggests your first sources. It’s just a directory of markdown, so you can open it in Obsidian, VS Code or any editor.

Want Hermes building a knowledge base for your business? Inside the AI Profit Boardroom you get the Agent OS — Hermes, OpenClaw, OpenCode, Claude and Codex in one place — plus step-by-step Hermes training and four live coaching calls a week.

Want a plan for your business first? Claim your free strategy session.

Join the AI Profit Boardroom →Claim Your Free SEO Strategy Session →

The three layers

Layer What it holds Who owns it
Raw Your original sources: articles, papers, transcripts and assets. Never edited. You
Wiki Summaries, entity and concept pages, comparisons and filed answers The agent
Schema SCHEMA.md: the rules for structure, naming and tags You and the agent together

Karpathy’s gist describes the same split and notes the schema can be a file like CLAUDE.md or AGENTS.md. In Hermes it’s SCHEMA.md. At the start of every session the agent reads SCHEMA.md, index.md and the last 20 to 30 log entries before it does anything, so it always knows the current state.

The folder structure

~/wiki/
  SCHEMA.md        # rules for the wiki
  index.md         # map of every page
  log.md           # append-only history (rotated)
  raw/             # articles, papers, transcripts, assets
  entities/        # people, companies, tools
  concepts/        # ideas and methods
  comparisons/     # side-by-side analysis
  queries/         # answers worth keeping

Raw files carry a little frontmatter — source URL, ingest date and a sha256 hash. If you re-ingest a source that hasn’t changed, the agent skips it; if it has changed, the agent flags the drift.

Ingest, query and lint

Ingest

  • The agent captures the source into raw/ (it can extract web pages and PDFs).
  • It talks through the key takeaways with you (skipped when running on a schedule).
  • It checks what already exists, then writes or updates pages, updates index.md and log.md, and reports the changes.

One source can update 5 to 15 pages. If an ingest would touch 10 or more existing pages, the agent asks first.

Query

  • It reads index.md (and searches once the wiki passes about 100 pages).
  • It reads the relevant pages and writes an answer that cites them with [[wikilinks]].
  • Valuable answers get filed in queries/ or comparisons/ and logged, so the wiki gets smarter from your questions.

Lint

  • Finds orphan pages, broken wikilinks and pages missing from the index.
  • Checks frontmatter, stale pages (over 90 days), contradictions and low-confidence claims.
  • Spots source drift, pages over about 200 lines, messy tags and a log that needs rotating (over 500 entries).
  • Reports issues by severity and logs the run.

Want Hermes running your research and knowledge base for you? Claim your free strategy session →

Claim Your Free SEO Strategy Session →

The conventions that keep it clean

  • Lowercase, hyphenated filenames.
  • YAML frontmatter on every page: title, created, updated, type, tags and sources, plus optional confidence, contested and contradictions.
  • At least two outbound [[wikilinks]] per page.
  • Provenance markers on pages built from three or more sources.
  • A tag taxonomy of 10 to 20 tags.
  • A page gets created when two or more sources mention something, or it’s central to one.
  • Split pages that grow past about 200 lines, and move retired pages to _archive/.

Obsidian integration and headless sync

The wiki works as an Obsidian vault out of the box: wikilinks are clickable, Graph View shows how everything connects, and the frontmatter powers Dataview queries. Set Obsidian’s attachment folder to raw/assets/. If you also use the obsidian skill, point OBSIDIAN_VAULT_PATH at the same folder as your wiki. Karpathy sums the pairing up neatly: “Obsidian is the IDE; the LLM is the program.”

If Hermes runs on a server, the skill describes syncing with Obsidian Headless. You need Node 22+ and an Obsidian Sync subscription:

npm install -g obsidian-headless
ob login
ob sync-create-remote --name "LLM Wiki"
ob sync-setup --vault "<vault-id>"
ob sync --continuous

You can run the continuous sync as a systemd user service so it keeps going after you log out. For more on pairing Hermes with Obsidian, see Hermes Obsidian.

LLM Wiki vs RAG

LLM Wiki RAG
How it works Compiles sources into a maintained wiki Searches raw documents at question time
Memory Persistent and compounding Rediscovers the answer each time
What you can see Readable markdown pages you can browse and edit Hidden chunks in a vector store
Scale An index works up to roughly 100 sources and hundreds of pages, per Karpathy Built for large document sets

RAG is still useful for huge document piles. For a focused research topic you return to again and again, the wiki wins because the work you put in stays done.

Troubleshooting the Hermes LLM wiki

  • The skill is missing. Bundled skills are copied into ~/.hermes/skills/ when you install and synced on hermes update. Restore it with hermes skills reset llm-wiki --restore.
  • The wiki is in the wrong place. Check WIKI_PATH in ~/.hermes/.env; without it the wiki goes to ~/wiki.
  • Answers ignore recent work. Make sure SCHEMA.md, index.md and log.md exist and are up to date — the agent reads them first.
  • Big ingests stall. Ingests touching 10+ pages wait for your approval.
  • You want a batch build. The skill mentions llm-wiki-compiler, a Node.js CLI that compiles a wiki in one pass; use the skill itself when you want the agent curating with you.

My verdict on the Hermes LLM wiki

It’s one of the most useful skills already sitting in your Hermes install. Set WIKI_PATH, run /llm-wiki, feed it a few good sources and lint it once a week, and you end up with a knowledge base that keeps getting better instead of a pile of chats you never read again. If you research anything seriously, set up the Hermes LLM wiki today.

About Julian Goldie

I’m Julian Goldie — founder of Goldie Agency and the AI Profit Boardroom. I test new AI agents almost every day for 432K+ YouTube subscribers and run Hermes, OpenClaw, OpenCode, Claude and Codex side by side in my own Agent OS. Claim your free strategy session or join the AI Profit Boardroom.

Want another angle? Read the AI Profit Boardroom guide to Hermes LLM wiki.

FAQ: hermes llm wiki

What is the Hermes LLM wiki?

A bundled Hermes Agent skill, llm-wiki, that builds and maintains a knowledge base of interlinked markdown pages based on Andrej Karpathy’s LLM Wiki pattern.

Where does Hermes store the wiki?

In the folder set by WIKI_PATH, for example in ~/.hermes/.env. If it isn’t set, the default is ~/wiki.

What are ingest, query and lint?

Ingest adds a source and updates pages, query answers questions from the wiki with citations, and lint health-checks the wiki for broken links, stale pages and contradictions.

Can I use the Hermes LLM wiki with Obsidian?

Yes. The wiki folder works as an Obsidian vault out of the box, with clickable wikilinks, Graph View and Dataview support.

How do I restore the llm-wiki skill if it’s missing?

Run hermes skills reset llm-wiki –restore. Bundled skills are also synced when you run hermes update.

Ready to put Hermes to work in your business? Join the AI Profit Boardroom for the Agent OS, setup training and daily updates.

Or claim your free strategy session and we’ll map out your setup.

Starting from zero? Get the free AI course and community.

Join the AI Profit Boardroom →Claim Your Free SEO Strategy Session →Get the FREE AI Course + Community →

Related reading

Last updated October 2026. This page is a living guide to hermes llm wiki — the facts here move fast and it is updated as they do.

Picture of Julian Goldie

Julian Goldie

Hey, I'm Julian Goldie! I'm an SEO link builder and founder of Goldie Agency. My mission is to help website owners like you grow your business with SEO!

Leave a Comment

WANT TO BOOST YOUR SEO TRAFFIC, RANK #1 & GET MORE CUSTOMERS?

Get free, instant access to our SEO video course, 120 SEO Tips, ChatGPT SEO Course, 999+ make money online ideas and get a 30 minute SEO consultation!

Just Click The Button Below To Get FREE, Instant Access!