---
title: 'Hermes Obsidian is WILD (FREE!)'
source: 'https://youtube.com/watch?v=ZXjw8vNCWAU'
video_id: 'ZXjw8vNCWAU'
date: 2026-11-19
duration_sec: 559
channel: 'Julian Goldie SEO'
---

# Hermes Obsidian is WILD (FREE!)

> Source: [Hermes Obsidian is WILD (FREE!)](https://youtube.com/watch?v=ZXjw8vNCWAU)

## Summary

This video demonstrates how to set up a shared memory system for AI agents using Obsidian, a free notes app. It addresses the problem of isolated agent memories and provides a step-by-step guide to creating a collaborative AI workflow.

### Key Points

- **The Problem of Isolated Memories** [00:03] — Each AI agent (Hermes, Claude Code, Codex, OpenClaw) has its own tiny memory, leading to a lack of collaboration and repeated mistakes.
- **Hermes' Memory Limitations** [00:49] — Hermes' built-in memory is tiny (2200 characters), and each agent remembers only for itself, like four helpers who never talk.
- **Obsidian as Shared Memory** [02:02] — Using Obsidian as a shared vault allows agents to read and write notes, creating a common memory. Andre Karpathy's quote: 'Obsidian is the IDE, the LLM is the programmer, the wiki is the codebase.'
- **Folder Structure for Organization** [02:31] — The PARA folder system (Projects, Areas, Resources, Archive) keeps the vault organized. Additional folders: inbox, daily, memories, and agent OS.
- **The Orchestrator Approach** [04:10] — Instead of configuring each agent by hand, pick one orchestrator (Claude Code or Codex) and give it a single prompt to set up everything, including a protocol note.
- **The Protocol Note** [04:59] — The protocol note contains all rules; changing it updates all agents. Each agent gets a pointer to read the vault and a nightly task to write summaries.
- **Voice Agent Integration** [05:29] — A voice agent can read notes back, and Hermes can run Instagram for mobile access. Example: asking 'What did my agents get done?'
- **The Mistake: Saving Everything** [06:00] — Saving every chat backfires due to token limits. Keep it lean: one summary per day with four parts (worked on, learned, decisions, in progress).
- **Extra Prompts for Efficiency** [07:18] — Extra prompts: morning brief (5 bullets), check before you start (review last 3 days), and health check (weekly).

## Transcript

each of your AI agents writes a short note about its day, what it worked on, what it finished, and what it learned. The next morning, any agent can read that note and pick up right where the others stopped. Hermes, Claude Code,
Codeex, and OpenClaw all read from one shared memory. You set the whole thing up with one prompt, and you never edit a file yourself. I'll show you the exact setup inside our agent OS. I'll also show you one mistake that makes agents
slower. And stick around for the part where my voice agent reads my notes back to me out loud. So, here's the problem. Hermes agent is the free open-source agent from Noose Research. It came out in February 2026, and it runs on your
own computer. It has memory, but the built-in memory is tiny on purpose. One file holds about 2200 characters, which is about one page in a kids book. The file about you holds even less. Plus, Hermes only remembers for Hermes. Claude
Code, Codeex, and OpenClaw each have their own separate memory. It's like four helpers in your office who never talk to each other. Every morning, each one starts with no idea what the others did yesterday. This is the memory galaxy
did yesterday. This is the memory galaxy inside our agent OS. Every dot is a note and every line is a link between two notes. There are thousands of memories in here. Click on Goldie Agency and you'll see who Julian is and what the
agency does, all linked together. Julian doesn't sort any of it. The agents keep it up to date. Obsidian is a free notes app. Your notes are plain text files in a folder on your own computer. Your agents can open and write them like any
other file. Andre Karpathy, one of the people who started OpenAI, shared a setup like this in April 2026. An AI agent keeps a folder of notes neat
and linked, and he uses Obsidian to look at it. His line was, "Obsidian is the at it. His line was, "Obsidian is the IDE, the LLM is the programmer, the wiki is the codebase." In plain words, Obsidian is the window, the AI does the
writing, and the notes are what gets built. Hermes users found this, too. A Reddit post about using obsidian as long-term memory for Hermes got close to 800 up votes. Hermes even comes with an obsidian skill built in. So, how do you
keep it from turning into a mess? You use folders. The folder system is called para projects, areas, resources, and archive. Projects have an end date like
launch the new AI profit boardroom welcome email. Areas never end, like the community or SEO. Resources are things you look up, like your brand voice.
Archive is where finish stuff goes. Quick pause before I show you the next part. If you want it, you can get the agent OS inside the AI profit boardroom. You get the full zip file ready to install with Hermes, the memory galaxy,
the voice agent, and the create section already set up. We've also built a already set up. We've also built a 30-day agent OS road map. Day one, you install the agent OS and connect Hermes. Day two, you set up your Obsidian Vault
with the right folders. Day three, you paste in the setup prompt and turn on nightly logs for every agent. There are four coaching calls every week where you can share your screen and get help live. A lot of members already run this exact
stack. link in the description or go to aiprofitboardroom.com. On top of that, we add four more folders. An inbox for messy notes, a daily folder for your own notes, a memories folder for decisions and wins,
and an agent OS folder that belongs to the agents. That last folder is how everything syncs. Claude, Hermes, and the voice agent all write into the same vault. Now, there's one choice that makes or breaks this, and most people
get it wrong. They set up each agent by hand, one at a time. Please don't do that. Instead, pick one agent to be the boss. We call it the orchestrator. Clawed code or codecs work best because both can read and write files all over
your computer. Then you give the boss one big prompt. Set up a shared Obsidian memory for all my agents. Do all the setup yourself. I don't want to edit these files by hand. Add your vault path and list your agents. For Julian, that's
Claude Code, Hermes, Codeex, and OpenClaw. The boss finds your agents, builds the folders, and writes one rules note called the protocol. It gives each agent its instructions, tests everyone, and reports back. Already have a vault?
It checks what's synced and fits the new system around what you have. When it's done, every agent gets two things. The first is a short pointer, so it reads the vault at the start of every session. The second is a nightly task, so it
writes its summary without you asking. All the rules live in that one protocol note. If you change the note, all four agents change. Now, the voice agent, Julian asks Hermes out loud, "Can you look at my vault and give me some ideas
of what we worked on?" Hermes answers from his notes. It tells him he'd been making video title ideas in Claude. He did that work in Claude. Then he asked a totally different agent by voice and Hermes knew. Hermes also runs Instagram
so you can ask from your phone. What did my agents get done on the AI profit boardroom yesterday? Now the mistake. You might think, why not save every chat and every message? It backfires. AI reads text in small
chunks called tokens, and each agent only has so much room. If you fill that room with junk, the useful stuff gets pushed out and answers get worse. People in our community have asked how to put millions of memories into Obsidian.
Honestly, you don't need that. Karpathy's wiki on one topic was about 100 articles. That was plenty because the agent kept it neat and linked. So, what does the nightly note look like? Each agent writes one summary for its
whole day. So, any agent can read it in under a minute. It has four parts. What it worked on, what it learned, decisions it made, and what's still in progress. Here's what one could look like. Hermes writes, "Worked on three subject lines
for the AI profit boardroom welcome email. Learned that the shortest one was the clearest. Pick that one. Still in progress. the day three follow-up. That learned line is the most important line in the note. Hermes already learns from
what it does, but that learning stays inside Hermes. Once Hermes writes it into the vault, Claude Code, Codeex, and Open Claw all learn it, too. So, if Hermes learns something on Monday, Claude Code uses it on Tuesday. Nobody
had to tell it. Once the logs are running, you can add a few extra prompts. Pick the ones that fit. The first is the morning brief. Read yesterday's logs. Send me five bullets covering what's done, what's stuck, and
my top three focus items. You wake up knowing where your AI profit boardroom projects stand. The second is check before you start. Check the last 3 days of memories for anything about this project, then carry on. So before Claude
code updates the AI profit boardroom landing page, it checks what Hermes already tried. The third is the health check. List each agent's nightly task, when it last ran, and fix any that failed. Nightly tasks fail sometimes, so
run this once a week. There's a window right now where most business owners haven't figured this out. They're still pasting the same background info into pasting the same background info into every agent every day. That window won't
stay open forever. If you want the exact system from this video, get the agent OS inside the AI profit boardroom. It's the full zip file ready to install. You get the memory galaxy, the voice agent, the create section, and the setup prompt
with the full Obsidian daily log guide. There are daily tutorials on using Hermes and the agent OS to save time and get more leads. There are four coaching calls every week where you can bring your vault and get help live. You get a
prompt library with the morning brief and the health check. And there's a member map so you can find people near you running Hermes and Obsidian 2. Link in the description or go to a profitboardroom.com.
And if you want the full process, the SOPs and over 100 AI use cases like this one, join the AI success lab. It's our free community links in the comments and free community links in the comments and description.
