summree
Build a Hermes Knowledge Base That Self-Improves
Hermes
Jack Roberts

Build a Hermes Knowledge Base That Self-Improves

⏱ 14 min video · 3 min read14 Jun 2026Worth watching
TL;DR
Jack Roberts shows how to extend the Hermes AI agent with a self-improving local knowledge base (an LLM Wiki / Obsidian RAG) inspired by Andrej Karpathy, so Hermes can access and query external knowledge like meeting notes, expert articles, and research — not just its own conversational memory. The result is a bidirectional system where Hermes knows the user personally and the wiki knows their world, combined into one unified AI operating system.
Key points
1
Hermes has strong conversational memory but a critical blind spot: it only knows what you have told it directly — not your inbox, meeting notes, research, or external expert knowledge.
2
Andrej Karpathy's LLM Wiki concept (also called Obsidian RAG) creates a self-referential, self-improving local knowledge base that rewrites itself as new content is added, flags contradictions, and grows indefinitely.
3
The wiki is set up by feeding the LLM Wiki URL to Claude (using Claude Opus 4 in the Claude app to avoid API costs) and having it scaffold the folder structure on your desktop, then connecting that folder path to Hermes.
4
You can manually ingest any article, transcript, or document into the wiki by pasting it into Hermes and asking it to index the content using the claude.md injection workflow, after which Hermes can query it in conversation.
5
Recurring automated tasks (cron jobs) can be set up inside Hermes to auto-ingest sources like Granola meeting transcripts daily, making the knowledge base grow passively without manual effort.
Actionable insights
Go to the LLM Wiki URL, paste it into Claude or Hermes, and ask it to create the desktop folder structure — this scaffolds your Obsidian Wiki automatically without manual setup.
Create a dedicated Hermes persona/skill (e.g. named 'LLM Wiki') with a system prompt telling Hermes when and how to query the wiki, so it consults external knowledge automatically during relevant conversations.
Set up a daily recurring task inside Hermes to pull new Granola meeting transcripts (or any calendar/note source) and index them into the wiki under a meeting notes section — zero ongoing manual effort required.
When you find a valuable article or expert content online, paste it directly into Hermes and say 'index this into my Obsidian Wiki following the claude.md principles' — it will be permanently queryable.
Because both Claude Code and Hermes Agent share access to the same local Obsidian Wiki folder, any knowledge added is instantly available across both tools without duplication.
Notable quotes

Hermes knows you, it doesn't know your inbox.

Hermes knows you, the wiki knows your world, and now when we combine this with Hermes Agent, your agent can read both and basically do very cool things when we wire these two things together.

Your knowledge base just grows exponentially and will just never forget the stuff that it needs to know.

Worth watching?
Worth watching the full video?
Worth watching if you actively use Hermes and want to see the exact setup steps live — the visual walkthrough of connecting the Obsidian Wiki folder, building the persona skill, and creating the cron job adds useful context beyond what the summary can convey.
Topics
AI & TechHermes

Explore more summaries on these topics →

Saved you some time? The creator still deserves a like.

Watch on YouTube →
More like this

Want this for your own channels?

Add the channels you follow. Every new video summarised and in your inbox the moment it drops. From £4/month.

Try it free