The Hermes LLM wiki is a free skill built into Hermes Agent that makes the agent build and maintain a knowledge base for you, as a folder of linked markdown pages that gets better every time you add a source.
It is based on the LLM Wiki pattern that Andrej Karpathy published on GitHub in April 2026, and Nous Research ships it with Hermes as the bundled llm-wiki skill.
You set it up by pointing the WIKI_PATH variable at a folder and typing /llm-wiki in Hermes, and the agent writes the starter files itself.
This page is about what a business actually does with it, so after a quick explainer I'll walk through six use cases and the exact operation each one relies on.
In the video below I walk through Karpathy's idea and the three operations, and I spend a chunk of it on who should care, including business teams and content creators.
Hermes LLM Wiki Use Cases At A Glance
This table is the short answer, and each use case gets its own section below.
| Use case | What you feed it | What you get back |
|---|---|---|
| 1. Competitor tracking | You feed it competitor pages, news and launch posts. | You get one page per competitor that stays current. |
| 2. A team wiki | You feed it meeting transcripts, project docs and call notes. | You get answers with links back to the exact pages. |
| 3. A content research vault | You feed it articles, papers and video transcripts. | You get topic pages that compound across every piece you make. |
| 4. Due diligence | You feed it a vendor's docs, terms and reviews. | You get a comparison page with contradictions flagged. |
| 5. Training and onboarding | You feed it course notes and internal how-tos. | You get a wiki a new hire can browse in Obsidian. |
| 6. A wiki per client | You feed each client's material into its own folder. | You get separate knowledge bases that never mix. |
Every one of these runs on the same three operations, which are ingest, query and lint.
🔥 Want my exact 6 Ways Businesses Use The Hermes LLM Wiki 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
What The Hermes LLM Wiki Is, In Plain English
Hermes Agent is the free, open-source AI agent from Nous Research.
The Hermes LLM wiki is one of the skills that ships with it, and the official skill page says it is bundled and installed by default.
The docs describe it as "Karpathy's LLM Wiki: build/query interlinked markdown KB."
Instead of searching your documents from scratch every time you ask something, the agent reads each source once and writes what it learned into wiki pages.
Those pages link to each other, flag contradictions and get updated when a new source comes in.
The docs sum up the division of labour clearly, because you curate the sources and direct the analysis while the agent summarises, cross-references, files and keeps everything consistent.
The wiki is just a folder of markdown files, so there is no database to run and nothing locks your knowledge into one tool.
Who Makes It And What It Costs
The idea comes from Andrej Karpathy's LLM Wiki gist, which he wrote as an idea file for people to paste into their own agent.
Nous Research turned it into the llm-wiki skill, version 2.1.0, released under the MIT licence for Linux, macOS and Windows.
The skill costs nothing, and Hermes Agent is free and open source.
The only running cost is the AI model you connect to Hermes, because that model does the reading, summarising and writing.
The Business Problem It Solves
Most businesses I talk to have the same problem with knowledge.
It lives in people's heads, in chat threads nobody scrolls back through and in documents nobody reopens.
Karpathy puts his finger on why company wikis die: "Humans abandon wikis because the maintenance burden grows faster than the value."
An agent doesn't get bored of the bookkeeping, so the Hermes LLM wiki keeps the cross-references, summaries and index current as part of every ingest.
That's the whole business case, because the expensive part of a knowledge base was never the writing, it was the upkeep.
How To Set Up The Hermes LLM Wiki In Five Minutes
You need Hermes Agent installed, and the official quickstart covers that.
- You run
hermes update, because the docs say every update syncs the bundled skills into your skills folder. - You set
WIKI_PATHin the.envfile in your Hermes home, or you leave it unset and the wiki goes in~/wiki. - You type
/llm-wikiin Hermes, because every installed skill is available as a slash command. - You tell the agent exactly what the wiki covers, and it writes
SCHEMA.md,index.mdandlog.md. - You give it your first source and let it build the first pages.
The folder it creates has a raw/ area for your untouched sources, plus entities/, concepts/, comparisons/ and queries/ for the pages the agent writes.
SCHEMA.md is the rulebook for naming, frontmatter and tags, so it's worth ten minutes of your time on day one.
If the skill ever goes missing or you break it, the documented hermes skills reset llm-wiki --restore command puts the original back.
Use Case 1: Competitor Tracking That Stays Current
Competitive analysis is one of the examples in Karpathy's own gist.
You feed the wiki each competitor's pricing page, launch posts and news coverage, and the agent creates an entity page per company.
When a competitor changes something, the next ingest updates the page and notes where the new information contradicts the old.
Hermes ships a bundled competitor-news-monitor skill, which the skills catalog describes as watching "named companies for material news" with cited digests.
The Hermes cron docs say a scheduled job can attach more than one skill by repeating the --skill flag, and their example is hermes cron create "every 1h" "Use both skills and combine the result" --skill blogwatcher --skill maps.
The same pattern lets you attach competitor-news-monitor and llm-wiki to one scheduled job, so the digest lands in the wiki instead of your inbox.
The llm-wiki skill expects this kind of automation, because it says the step where it discusses takeaways with you is skipped in automated or cron contexts.
Use Case 2: A Team Wiki That Maintains Itself
Karpathy's gist describes an internal wiki fed by Slack threads, meeting transcripts, project documents and customer calls, with humans reviewing the updates.
The Hermes folder structure even has a raw/transcripts/ folder for meeting notes and interviews.
When someone asks a question, the agent reads index.md, opens the relevant pages and answers with citations like "Based on [[page-a]] and [[page-b]]."
If the answer is a substantial comparison or analysis, it gets filed into queries/ or comparisons/, so the next person who asks gets it instantly.
The human review is built in, because the skill tells the agent to confirm with you before any ingest that would touch 10 or more existing pages.
If you're working out how agents like this could run research inside your agency, you can book a free SEO strategy session and my team will talk it through with you.
🔥 Want the exact Hermes setup I use for research? Inside the AI Profit Boardroom, I've got a full Hermes Agent section with step-by-step video tutorials and the installable Agent OS. You'll be learning alongside 3,400+ members with four coaching calls every week. → Get access here
Use Case 3: A Content Research Vault That Compounds
This is the one I'd start with if you make videos, posts or newsletters.
In my video I call it a research vault that compounds over time instead of starting fresh on every topic.
You drop articles, papers and transcripts into the wiki as you research, and the agent builds concept pages that grow with every piece you make.
Karpathy recommends the Obsidian Web Clipper browser extension to turn web articles into markdown quickly.
Next month, when you write about a related topic, the groundwork is already sitting in the wiki with links to every source.
Pages that pull from three or more sources carry provenance markers such as ^[raw/articles/source-file.md], so you can trace a claim back before you publish it.
I've covered how Hermes fits SEO work more broadly in my post on Hermes Agent for SEO.
Use Case 4: Due Diligence Before You Buy Or Partner
Due diligence is another example straight from Karpathy's list.
You ingest a vendor's documentation, terms and public reviews, and then ask the wiki for a side-by-side comparison.
The skill's template says comparison pages should state what is compared and why, lay out the dimensions in a table and end with a verdict.
When two sources disagree, the update policy tells the agent to note both positions with dates and mark the page with contradictions in its frontmatter.
Pages can also carry an optional confidence field, and the lint pass lists low-confidence pages so weak claims don't quietly become accepted fact.
That's exactly the discipline you want before signing a contract.
Use Case 5: Training And Onboarding New Staff
Course notes are on Karpathy's list too, and they map neatly onto onboarding.
You ingest your internal how-tos, recorded training notes and process documents, and the agent builds concept pages for each process.
The docs say the wiki works as an Obsidian vault out of the box, with clickable wikilinks, Graph View and Dataview queries.
A new hire can open the folder in Obsidian, click through the graph and see how every process connects.
If your team works on a server, the skill documents Obsidian Headless, which syncs the vault through Obsidian Sync and needs Node.js 22 or newer plus a Sync subscription.
Use Case 6: A Separate Wiki For Every Client
Agencies need client knowledge kept apart, and the Hermes LLM wiki handles that with one setting.
The skill reads WIKI_PATH from the .env file in your Hermes home, and the Hermes docs say each profile has its own Hermes home.
So a Hermes profile per client can point its WIKI_PATH at that client's own wiki folder.
Each client then gets its own schema, its own tag taxonomy and its own log, and nothing leaks from one knowledge base into another.
I'd still keep sensitive client material on machines you control, because the raw sources are stored as plain files.
The Weekly Habit That Keeps It Valuable
Every use case above decays without a regular lint pass.
You ask Hermes to lint the wiki, and it checks for orphan pages, broken wikilinks, pages missing from the index, stale pages, contradictions and source drift.
Stale means a page whose updated date is more than 90 days older than the newest source on the same subject.
It also flags pages over 200 lines for splitting, and it rotates log.md once it passes 500 entries.
The report comes back grouped by severity, so you fix broken links first and style issues last.
The Limits To Know Before You Promise Anyone Anything
| Limit | What it means for a business |
|---|---|
| The skill is instructions, not a model. | Your results depend on the model you connect to Hermes. |
| The index has a sweet spot. | Karpathy says it works well at around 100 sources and hundreds of pages, and the skill adds file search past 100 pages. |
| Big updates wait for you. | Any ingest that would touch 10 or more existing pages needs your approval. |
| Raw files are read, never edited. | Corrections go into wiki pages, so your originals stay intact. |
| Sources sit on disk as plain files. | You decide where the folder lives and who can reach it. |
There is also a Node.js CLI called llm-wiki-compiler that the docs mention for batch compiling a source folder.
The docs say it owns page generation and is tuned for small corpora, so the Hermes skill is the better fit when you want the agent curating with you.
If you want to see where this sits among other ways Hermes remembers things, my Hermes Agent memory roundup compares the options, and my list of Hermes Agent use cases shows what else businesses run on it.
🔥 Want help turning Hermes into your team's research engine? The AI Profit Boardroom has Hermes Agent training, the Agent OS and four weekly coaching calls where you can ask me how to set up a knowledge base for your business. → Join the AI Profit Boardroom
Also On Our Network
- 🌐 The step-by-step Hermes LLM wiki setup for operators
- 🌐 Every Hermes LLM wiki feature, reviewed and scored
- 🌐 How the Hermes LLM wiki fits an Agent OS workflow
- 🌐 The models that run a Hermes LLM wiki best, on GoldieBench
FAQ: Hermes LLM Wiki For Business
What is the Hermes LLM wiki?
The Hermes LLM wiki is the bundled llm-wiki skill in Hermes Agent from Nous Research.
It makes the agent build and maintain a persistent knowledge base of linked markdown pages, based on Andrej Karpathy's LLM Wiki pattern.
Can a business use the Hermes LLM wiki as a team wiki?
Yes, and Karpathy's own gist lists an internal team wiki fed by Slack threads, meeting transcripts, project documents and customer calls as an example.
The Hermes skill asks before any ingest that would touch 10 or more existing pages, so a human stays in the loop.
Does the Hermes LLM wiki cost anything?
The skill is free and MIT-licensed, and Hermes Agent itself is free and open source.
Your running cost is the AI model you connect to Hermes, unless you run a local model.
Can the Hermes LLM wiki update itself on a schedule?
Hermes cron jobs can attach one or more skills with the --skill flag.
The llm-wiki skill says it skips the discuss-takeaways step in automated or cron contexts, so scheduled ingests are an expected way to run it.
How big can a Hermes LLM wiki get?
Karpathy says an index file works surprisingly well at around 100 sources and hundreds of pages.
The Hermes skill adds a file search once a wiki passes 100 pages and splits the index once sections pass 50 entries.
If you'd like a plan for using agents like this to grow your business, book a free SEO strategy session and we'll map it out together.
📺 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
Pick one use case, feed it one good source a day and lint it every Friday, and your business will have a knowledge base that finally maintains itself with the Hermes LLM wiki.
