agentos.guide › blog

Hermes LLM Wiki: The Quick Start For Karpathy's Wiki Skill, And How It Fits Your Agent OS

By Julian Goldie · 27 October 2026 · agentos.guide

The Hermes LLM wiki is the bundled llm-wiki skill in Hermes Agent, which makes the agent build and maintain a knowledge base as a folder of linked markdown pages.

It is based on Andrej Karpathy's LLM Wiki pattern, and Nous Research ships it with Hermes, installed by default.

You set WIKI_PATH to a folder, type /llm-wiki, and Hermes writes the schema, index and log, then ingests, queries and lints the wiki for you.

This page is the quick start, plus how it fits an Agent OS that runs several agents side by side.

The Hermes LLM wiki quick start

This is the shortest path from a fresh Hermes install to a working wiki.

Every step comes from the official Llm Wiki skill page and the Hermes skills docs.

If WIKI_PATH isn't set, the skill falls back to ./wiki, using the line WIKI="${WIKI_PATH:-$HOME/wiki}".

The docs say every installed skill becomes a slash command, so /llm-wiki works in the CLI and on any messaging platform Hermes is connected to.

You can also preload it for a whole session, because hermes chat takes -s or --skills to preload skills.

What the Hermes LLM wiki is

Hermes Agent is the free, open-source AI agent from Nous Research.

The Hermes LLM wiki is its bundled llm-wiki skill, which the docs describe as "Karpathy's LLM Wiki: build/query interlinked markdown KB."

The docs list it as installed by default, version 2.1.0, MIT-licensed, for Linux, macOS and Windows, with obsidian and arxiv as related skills.

When I checked my own Hermes install, which runs v0.22.1, the skill was already in my skills folder without me installing anything.

The skill costs nothing, and the only running cost is the model you connect to Hermes.

Where the idea comes from

Andrej Karpathy published the pattern as a GitHub gist called LLM Wiki in April 2026.

He wrote it as an idea file to paste into an agent such as Codex, Claude Code or OpenCode, and he kept it deliberately abstract.

His argument is that RAG-style tools rediscover knowledge on every question, while an agent-maintained wiki compiles it once and keeps it current.

He splits the work cleanly, with the human curating sources and asking questions, and then he adds: "The LLM's job is everything else."

The git history of the Hermes repo shows Nous added the skill two days after the gist went up.

In my video below I walk through the idea, the three layers and the three operations.

The folder Hermes builds for you

On a new wiki, the skill tells Hermes to create this structure, ask you what the wiki covers and write the three starter files.

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 raw/ layer holds your sources, and the agent reads them but never edits them.

The pages in entities/, concepts/, comparisons/ and queries/ are written and maintained by the agent.

SCHEMA.md is the rulebook for filenames, frontmatter and the 10 to 20 tags the wiki is allowed to use.

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.

🔥 Want the exact Hermes Agent setup I run inside my Agent OS?

Inside the AI Profit Boardroom you get the installable Agent OS, step-by-step video tutorials, four coaching calls a week and 3,400+ members building real automations.

→ Get access here

Your first session, step by step

The skill gives Hermes a fixed order of work for a brand-new wiki, so you know exactly what to expect.

• First, it works out the wiki path from WIKI_PATH, or asks you, with ~/wiki as the default.
• Next, it creates the folder structure shown above.
• Then it asks you what domain the wiki covers, and it wants a specific answer rather than a vague one.
• After that, it writes a SCHEMA.md customised to your domain, an index.md with section headers and a log.md with a creation entry.
• Finally, it confirms the wiki is ready and suggests the first sources to ingest.

My advice is to answer the domain question narrowly, because a tight domain gives you a tight tag taxonomy and cleaner pages.

The schema's page thresholds mean a page only gets created when something appears in two or more sources or is central to one.

That rule stops the wiki filling up with pages about passing mentions.

The rules that keep a wiki healthy

The skill ends with a pitfalls list, and these are the rules worth remembering.

• The agent never modifies files in raw/, because sources are immutable and corrections go into wiki pages.
• Every new page links to at least two other pages, because isolated pages are invisible.
• Every tag comes from the taxonomy in SCHEMA.md, and new tags are added there first.
• Contradictions are never silently overwritten, because the agent notes both claims with dates and flags them for review.
• Pages over about 200 lines get split, and fully superseded pages move to an _archive/ folder.

The three operations you'll use every day

Each raw source gets a sha256 hash in its frontmatter, so a re-ingest skips unchanged sources and flags drift.

Pages built from three or more sources carry provenance markers such as ^[raw/articles/source-file.md], so you can trace each claim.

The optional confidence and contested fields let lint surface weak claims before they harden into accepted fact.

Lint also flags pages that are more than 90 days stale, pages over 200 lines and a log that has passed 500 entries.

🔥 Want the exact Hermes Agent setup I run inside my Agent OS?

Inside the AI Profit Boardroom you get the installable Agent OS, step-by-step video tutorials, four coaching calls a week and 3,400+ members building real automations.

→ Get access here

How the Hermes LLM wiki fits an Agent OS workflow

My Agent OS runs several agents side by side, and the hard part is always shared knowledge rather than raw intelligence.

The cron docs show the multi-skill pattern as hermes cron create "every 1h" "Use both skills and combine the result" --skill blogwatcher --skill maps, and you swap in the skills you need.

A natural pairing is the bundled competitor-news-monitor skill, which the catalog says watches named companies for material news with cited digests.

If you want the other direction, where other agents query Hermes, my guide to running Hermes as an MCP server covers that.

Viewing your wiki in Obsidian

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

They recommend setting Obsidian's attachment folder to raw/assets/.

If you use the bundled obsidian skill too, set OBSIDIAN_VAULT_PATH to the same folder as your wiki.

On a server with no display, the skill documents npm install -g obsidian-headless, which needs Node.js 22 or newer and an Obsidian Sync subscription.

After ob login, ob sync-create-remote --name "LLM Wiki" and ob sync-setup --vault "<vault-id>" inside the wiki folder, ob sync --continuous keeps it syncing.

I built a vault-based second brain for Hermes in my Hermes second brain guide, and the Obsidian habits carry straight over.

Common Hermes LLM wiki problems and fixes

The docs also mention llm-wiki-compiler, a Node.js CLI for batch-compiling a source folder into a wiki.

They say it owns page generation and is tuned for small corpora, so the skill is the better fit for a wiki you keep curating.

The limits to plan around

The skill is a set of instructions, so the quality of every page depends on the model you connect.

Karpathy says an index file works well at around 100 sources and hundreds of pages, and the skill adds file search past 100 pages.

Big ingests wait for approval, which is safe but slows fully unattended runs.

Your sources sit on disk as plain files, so keep the folder somewhere you control.

If you want to talk through where an agent knowledge base fits an SEO business, you can book a free SEO strategy session with my team.

FAQ

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 linked markdown files, based on Andrej Karpathy's LLM Wiki pattern, and it is installed by default.

Who created the LLM Wiki pattern?

Andrej Karpathy published it as an idea file in a GitHub gist in April 2026. Nous Research turned it into the llm-wiki skill that ships with Hermes Agent.

How do I start a Hermes LLM wiki?

Set WIKI_PATH in the .env file in your Hermes home, or leave it unset to use ~/wiki, then type /llm-wiki and tell Hermes what the wiki covers.

Can the Hermes LLM wiki run on a schedule?

Yes. Hermes cron jobs can attach skills with --skill, and the llm-wiki skill skips the discuss-takeaways step in automated or cron contexts.

Does the Hermes LLM wiki need a database?

No. The docs say it is just a directory of markdown files, so you can open it in Obsidian, VS Code or any editor.

How do I restore the llm-wiki skill?

Run hermes skills reset llm-wiki --restore, which deletes your local copy and re-copies the current bundled version.

Final word on the Hermes LLM wiki

Set the folder, run the slash command, feed it one good source at a time and lint it weekly, and your Agent OS gets a research memory that compounds.

If you want a second opinion on how agents should fit your SEO business, you can book a free SEO strategy session with my team.

If you want the whole Agent OS with Hermes already wired in, it lives inside the AI Profit Boardroom, with 3,400+ members and four coaching calls a week.

That is the fastest way to get real value from the Hermes LLM wiki.

Related guides on agentos.guide

→ The Hermes Second Brain

→ Hermes /learn: The Book Brain Engine

→ Hermes AI Skills

→ The Hermes Crash Course

📺 Video notes + links to the tools 👉

🎥 Learn how I make these videos 👉

🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉

← all guides on agentos.guide