The Hermes LLM wiki is the bundled llm-wiki skill in Hermes Agent, which turns Hermes into an agent that builds and maintains its own knowledge base as a folder of linked markdown pages.
The pattern comes from AI researcher Andrej Karpathy, who published it as his LLM Wiki idea file on GitHub in April 2026.
Nous Research turned that idea into a skill, and the official docs say it ships with Hermes and is installed by default.
To set it up, you point the WIKI_PATH variable at a folder, run /llm-wiki in Hermes, and let the agent create the schema, index and log for you.
This page is the operator walkthrough, so every step below comes with the exact command or file layout from the official Hermes docs.
In the video below I walk through Karpathy's idea, the three layers and the three core operations, which is the best primer before you touch a terminal.
Hermes LLM Wiki Setup At A Glance
This table is the whole setup on one screen, and each step has its own section further down.
| Step | What you do | Command or file |
|---|---|---|
| 1 | You confirm the bundled skill is installed. | hermes skills list |
| 2 | You choose where the wiki lives. | WIKI_PATH in your Hermes .env file |
| 3 | You start the wiki from chat. | /llm-wiki |
| 4 | You tell the agent what the wiki covers. | SCHEMA.md, index.md and log.md get written |
| 5 | You feed it your first source. | A URL, a PDF or pasted text |
| 6 | You ask it questions. | Answers cite [[wikilinks]] |
| 7 | You run a health check. | "Lint my wiki" |
The source for all of this is the official Llm Wiki skill page in the Hermes docs.
What The Hermes LLM Wiki Actually Is
Hermes Agent is the free, open-source AI agent from Nous Research.
The Hermes LLM wiki is one of its bundled skills, listed in the skills catalog as llm-wiki under research.
The docs describe it as "Karpathy's LLM Wiki: build/query interlinked markdown KB."
This is the skill metadata as the docs show it.
| Field | Value |
|---|---|
| Source | It is bundled and installed by default. |
| Path | It lives at skills/research/llm-wiki. |
| Version | The current version is 2.1.0. |
| Licence | It is released under the MIT licence. |
| Platforms | It runs on Linux, macOS and Windows. |
| Related skills | The related skills are obsidian and arxiv. |
When I checked my own Hermes install, which runs v0.22.1, the skill was already sitting in my skills folder at version 2.1.0 without me installing anything.
The git history of the NousResearch/hermes-agent repo shows the skill was added in April 2026, a couple of days after Karpathy's gist went up.
🔥 Want my exact Hermes LLM Wiki Setup setup, step by step? Inside the AI Profit Boardroom you get step-by-step video trainings, my installable Agent OS and four coaching calls a week with 3,400+ members building real automations. → Join the AI Profit Boardroom
Where The Idea Comes From: Karpathy's LLM Wiki
Karpathy's original is a GitHub gist called LLM Wiki, and he describes it as a pattern for building personal knowledge bases with LLMs.
He calls it an idea file that you paste into your own agent, such as Codex, Claude Code or OpenCode.
His point is that RAG-style tools rediscover the same knowledge on every question, so he has the model build a persistent wiki that sits between you and your raw sources instead.
He keeps Obsidian open on one side and the agent on the other, and he sums the setup up neatly: "Obsidian is the IDE; the LLM is the programmer; the wiki is the codebase."
The gist is deliberately abstract, and the Hermes skill fills that gap with a concrete folder structure, schema template and set of rules.
The Problem It Solves
Every chat with an agent is useful for a few minutes, and then the research disappears into history.
The Hermes LLM wiki fixes that by compiling what you read into pages that stay, so the second question starts where the first one finished.
In the docs' words, the human curates sources while the agent summarises, cross-references, files and maintains consistency.
Before You Start
You need a working Hermes Agent install, and the official quickstart covers that for every platform.
You also need a model connected, because the skill is instructions and the model does the reading and writing.
You should update Hermes first, because the docs say each hermes update syncs the bundled skills into your skills folder.
If you set up a profile with no bundled skills, the docs say a .no-bundled-skills marker blocks seeding, and hermes skills opt-in --sync removes the marker and re-seeds.
Step 1: Confirm The Skill Is Installed
Run the skills list and look for llm-wiki in the research category.
hermes skills list
The docs say all skills live in ~/.hermes/skills/, and that bundled skills are copied there from the repo on a fresh install.
If it is missing or you edited it and want the original back, the documented reset command restores the pristine bundled copy.
hermes skills reset llm-wiki --restore
The same reset also works inside chat as /skills reset llm-wiki --restore, because every hermes skills command has a /skills version.
Step 2: Choose Where Your Wiki Lives
The skill reads the WIKI_PATH environment variable, and the docs suggest setting it in the .env file in your Hermes home.
If you leave it unset, the wiki goes in ~/wiki.
This is the exact line the skill uses to resolve the folder.
WIKI="${WIKI_PATH:-$HOME/wiki}"
🔥 Want my Hermes setup with the skills already wired in? Inside the AI Profit Boardroom, I've got a full Hermes Agent course plus the installable Agent OS, which runs Hermes, Claude and other agents under one shared memory. You'll be learning alongside 3,400+ members with four coaching calls every week. → Get access here
Step 3: Start The Wiki With /llm-wiki
The Hermes docs say every installed skill is automatically available as a slash command, so you type /llm-wiki in the CLI or any messaging platform.
You don't even need the slash command, because the skill activates when you ask to create a wiki, add a source, ask a question while a wiki exists, lint the wiki or mention your notes in a research context.
You can also preload it for a whole session, because hermes chat takes -s or --skills to preload one or more skills.
Step 4: Let Hermes Build The Scaffolding
On a new wiki, the skill tells the agent to follow seven steps in order.
- It works out the wiki path from
WIKI_PATH, or asks you, with~/wikias the default. - It creates the folder structure.
- It asks you what domain the wiki covers, and it wants a specific answer.
- It writes a
SCHEMA.mdcustomised to that domain. - It writes an initial
index.mdwith section headers. - It writes an initial
log.mdwith a creation entry. - It confirms the wiki is ready and suggests your first sources.
This is the folder structure it creates, copied from the docs.
wiki/
├── SCHEMA.md # Conventions, structure rules, domain config
├── index.md # Sectioned content catalog with one-line summaries
├── log.md # Chronological action log (append-only, rotated yearly)
├── raw/ # Layer 1: Immutable source material
│ ├── articles/ # Web articles, clippings
│ ├── papers/ # PDFs, arxiv papers
│ ├── transcripts/ # Meeting notes, interviews
│ └── assets/ # Images, diagrams referenced by sources
├── entities/ # Layer 2: Entity pages (people, orgs, products, models)
├── concepts/ # Layer 2: Concept/topic pages
├── comparisons/ # Layer 2: Side-by-side analyses
└── queries/ # Layer 2: Filed query results worth keeping
The three layers in plain English
The raw layer holds your original sources, and the agent reads them but never edits them.
The wiki layer holds the pages the agent writes, which are entity pages, concept pages, comparisons and filed answers.
The schema layer is SCHEMA.md, which defines the structure, conventions and tag taxonomy.
Karpathy's gist suggests a file like CLAUDE.md or AGENTS.md for the schema, and the Hermes skill uses a dedicated SCHEMA.md instead.
Spend ten minutes on the schema
The schema is the rulebook, so a vague domain gives you a vague wiki.
The template asks for lowercase hyphenated filenames, YAML frontmatter on every page and at least two outbound [[wikilinks]] per page.
It asks you to define 10 to 20 top-level tags, and every tag on a page must come from that list.
Step 5: Ingest Your First Source
Give Hermes a URL, a PDF or a block of pasted text and ask it to add it to the wiki.
The skill runs six steps for every ingest.
- It captures the raw source, using
web_extractfor URLs and PDFs, and saves it underraw/. - It discusses the key takeaways with you, unless it is running in an automated or cron context.
- It checks
index.mdand searches for existing pages so it doesn't create duplicates. - It writes new pages or updates existing ones, with cross-links and tags from your taxonomy.
- It updates
index.mdand appends an entry tolog.md. - It reports every file it created or changed.
The docs say a single source can update 5 to 15 wiki pages, and they call that the compounding effect.
If an ingest would touch 10 or more existing pages, the skill tells the agent to confirm the scope with you first.
Each raw file gets frontmatter with the source URL, the ingest date and a sha256 hash, so a re-ingest skips unchanged sources and flags drift.
🚀 Ready to put Hermes LLM Wiki Setup to work in your business? Inside the AI Profit Boardroom you get step-by-step video trainings, my installable Agent OS and four coaching calls a week with 3,400+ members building real automations. → Join the AI Profit Boardroom
Step 6: Ask Your Wiki Questions
Once a few sources are in, you just ask Hermes questions about the topic.
The agent reads index.md first, and on wikis with more than 100 pages it also searches across every markdown file.
It reads the relevant pages and writes an answer that cites them, such as "Based on [[page-a]] and [[page-b]]."
If the answer is a substantial comparison or deep dive, it files it back into queries/ or comparisons/ so you never have to re-derive it.
Step 7: Lint The Wiki Every Week
Linting is the health check, and you start it by asking Hermes to lint, audit or health-check your wiki.
The skill lists 13 lint steps, and these are the checks that matter most in practice.
| Check | What it finds |
|---|---|
| Orphan pages | It finds pages that no other page links to. |
| Broken wikilinks | It finds links that point at pages that don't exist. |
| Stale content | It finds pages more than 90 days older than the newest source on the same entities. |
| Contradictions | It surfaces pages marked contested or listing contradictions. |
| Source drift | It recomputes raw file hashes and flags mismatches. |
| Page size | It flags pages over 200 lines as candidates for splitting. |
The report is grouped by severity, with broken links first, and the run gets logged.
If you want help turning a wiki like this into a research engine for an SEO business, you can book a free SEO strategy session with my team.
Opening Your Hermes LLM Wiki In Obsidian
The docs say the wiki folder works as an Obsidian vault out of the box.
Wikilinks are clickable, Graph View shows how the pages connect, and the YAML frontmatter powers Dataview queries.
For the best results, the docs say to set Obsidian's attachment folder to raw/assets/ and install the Dataview plugin.
If you also use the bundled obsidian skill, the docs say to set OBSIDIAN_VAULT_PATH to the same folder as your wiki path.
I've written about pairing Claude with a vault in my Claude and Obsidian second brain setup, and the same habits carry straight over to Hermes.
Running it on a server with Obsidian Headless
If Hermes runs on a server with no display, the skill documents obsidian-headless, which syncs the vault through Obsidian Sync without the desktop app.
It needs Node.js 22 or newer and an Obsidian account with a Sync subscription.
These are the setup commands exactly as the docs show them, with your own details in place of the placeholders.
# Requires Node.js 22+
npm install -g obsidian-headless
# Login (requires Obsidian account with Sync subscription)
ob login --email <email> --password '<password>'
# Create a remote vault for the wiki
ob sync-create-remote --name "LLM Wiki"
# Connect the wiki directory to the vault
cd ~/wiki
ob sync-setup --vault "<vault-id>"
# Initial sync
ob sync
# Continuous sync (foreground — use systemd for background)
ob sync --continuous
The docs also give a systemd user service so the sync keeps running in the background after you log out.
🔥 Want help building your own agent knowledge base? The AI Profit Boardroom has Hermes Agent training, the Agent OS and four weekly coaching calls where you can ask me how to structure your wiki, your schema and your sources. → Join the AI Profit Boardroom
Troubleshooting The Hermes LLM Wiki
These are the problems the docs themselves point to, with the fix for each.
| Problem | Why it happens | Fix |
|---|---|---|
| The skill doesn't appear | Your profile may have opted out of bundled skills. | Run hermes skills opt-in --sync to re-seed them. |
| Updates never reach the skill | Hermes treats an edited bundled skill as user-modified and skips it. | Run hermes skills reset llm-wiki to re-baseline it, or add --restore for the original. |
| The wiki is in the wrong folder | WIKI_PATH isn't set, so it fell back to ~/wiki. |
Set WIKI_PATH in your Hermes .env file. |
| The agent duplicates pages | It skipped orientation at the start of the session. | Ask it to read SCHEMA.md, index.md and the recent log first. |
| A big ingest pauses | The ingest would touch 10 or more existing pages. | Approve the scope, and the agent carries on. |
The docs also mention llm-wiki-compiler, a Node.js CLI with the same Karpathy inspiration that batch-compiles a source folder into a wiki.
They say it owns page generation and is tuned for small corpora, so you use the skill when you want the agent curating with you.
What My Video Covers And What Changed Since
In my video I cover the core idea, the three layers, ingest, query and lint, and who the pattern suits.
The skill has moved on since it first landed, because a later commit ported provenance markers, source hashing and quality signals from llm-wiki-compiler into it.
Provenance markers trace each claim to a raw file, and the optional confidence and contested fields let lint surface weak claims for review.
If you want a wider view of how Hermes remembers things beyond this skill, my Hermes Agent memory roundup compares the options, and my best Hermes Agent skills list shows where llm-wiki fits.
Also On Our Network
- 🌐 Six ways a business can use the Hermes LLM wiki
- 🌐 The Hermes LLM wiki reviewed feature by feature
- 🌐 A Hermes LLM wiki quick start for your Agent OS
- 🌐 Which model should maintain your wiki, measured on GoldieBench
FAQ: Hermes LLM Wiki
What is the Hermes LLM wiki?
The Hermes LLM wiki is the bundled llm-wiki skill in Hermes Agent.
It builds and maintains a persistent knowledge base of interlinked markdown files, based on Andrej Karpathy's LLM Wiki pattern, and it is installed by default.
Who came up with the LLM Wiki idea?
Andrej Karpathy published the LLM Wiki pattern as an idea file in a GitHub gist in April 2026.
The Hermes skill links to that gist and turns the idea into instructions the agent follows.
Where does Hermes save the wiki?
Hermes uses the folder in the WIKI_PATH environment variable, which the docs suggest setting in the .env file in your Hermes home.
If WIKI_PATH is not set, the wiki goes in ~/wiki.
Do I need a database or vector store for the Hermes LLM wiki?
No, because the docs say the wiki is just a directory of markdown files with no database and no special tooling.
You can open it in Obsidian, VS Code or any editor.
How do I get the llm-wiki skill back if I edited or lost it?
Bundled skills are synced into your Hermes skills folder on install and on every hermes update.
The documented hermes skills reset llm-wiki --restore command deletes your local copy and re-copies the current bundled version.
If you're planning to put agents like this to work across your business, book a free SEO strategy session and my team will help you map it out.
📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉
About Julian
I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
- I've built a 7-figure SEO and link-building agency, Goldie Agency, with 70+ people in Bangkok.
- I've grown a YouTube channel to 400,000+ subscribers.
- I wrote the book "Link Building Mastery", which covers nine link building strategies.
- I run GoldieBench, where I score AI models on real one-shot builds.
→ Get my best AI training inside the AI Profit Boardroom
Set WIKI_PATH, run /llm-wiki, feed it one good source a day and lint it every week, and you'll have a knowledge base that keeps getting smarter with the Hermes LLM wiki.
