What Hermes Obsidian memory is and how it works
Hermes Obsidian memory is a shared notebook for all your AI agents. Each agent writes a short note about its day every night. The next morning, any agent can read that note and continue where others left off.
Obsidian is a free notes app. Your notes are plain text files stored on your computer. Plain text is easy for AI agents to read and write. Any agent that can open a file can use this.
In this episode, Julian Goldie explains how this works with Hermes, Claude Code, Codex, and OpenClaw. Every dot in the memory galaxy is a note. Every line is a link between notes. You click on a note like Facebook and it links to your funnel plan and about me note.
The best part is you don't keep it tidy. Your AI agents update the notes for you.
Main features of the shared memory system
The system has several key features. First, each agent gets a pointer in its instructions. It says where the vault is and to read it when a new chat starts. Second, each agent gets a nightly task to write its summary.
- Daily logs: One summary per day per agent, not one per task.
- Learned line: The most important part. Lessons from one agent go to all agents.
- One rulebook: Change one note and every agent follows the new rule.
- Agent OS folder: A place where agents talk to each other, with spots for Claude, Hermes, and others.
The orchestrator sets everything up. It reads your vault, checks your agents, builds folders, writes a rulebook, adds instructions to each agent, tests each one, and gives you a report.
It works on a fresh setup or an existing vault. The model checks what you already have and fits the new system in.
How to set up Hermes Obsidian memory step by step
Setting up this memory system is simple. You use one mega prompt with a boss agent, called the orchestrator. Julian recommends using Claude Code or Codex because they can read and write files all over your computer.
- Change the vault path to where your Obsidian vault lives on your computer.
- List the agents you use, like Claude Code, Hermes, Codex, and OpenClaw.
- Paste the mega prompt into a new chat with the orchestrator and hit enter.
- Let the orchestrator look at your vault, check your agents, and build the folders.
- Wait for it to write a rulebook and add a pointer note to every other agent.
- Let it test each agent to make sure everything works.
- Read the final report to see the vault, each agent's role, the test results, and the rulebook.
- Token budget: Too many notes mean more reading, which makes agents slow and confused and burns tokens fast.
- Keep notes short: Write one daily log per agent, not one per task.
- Use folders: The PARA system (Projects, Areas, Resources, Archives) plus inbox, daily, memories, and agent OS keeps thousands of notes organized.
- Don't skip the agent list: The orchestrator needs to know who to set up.
- Don't edit files by hand: Let the orchestrator do all the tricky bits.
Don't skip listing your agents. The orchestrator needs to know who to set up. You never edit files by hand. The orchestrator does all the setup itself.
How to use the shared memory in daily work
Once the memory system is set up, using it is easy. Each agent writes one short summary at the end of the day. The summary has five parts: worked on, finished, learned, decisions, and still in progress.
For example, Hermes Apollo, a voice agent, can look at your vault. Julian asks it, "Can you look at my vault and give me some ideas of what we worked on?" The agent checks its notes and tells him out loud what it has been doing.
This works in the voice agent, in Hermes chat, and in any agent hooked up to the vault. The learned line is the most important. When Hermes teaches itself something and writes it in learned, Claude Code reads it the next morning. So does Codex and OpenClaw.
One agent learns something and the whole team gets smarter. You don't have to ask any agent to write its summary. The nightly task does it automatically.
Results and tests shown in the episode
Julian demonstrates the system live on screen. He shows the memory galaxy with thousands of linked dots. He clicks on Facebook and it links to the funnel plan and About Me note. He clicks on Goldie Agency and it shows all the details about him.
He also shows the orchestrator in action. He starts a new chat in Claude, pastes the mega prompt, and the agent gets to work. It looks at the vault, checks which agents are installed, builds folders, writes a rulebook, adds notes to each agent, tests each one, and gives a report.
He tests the voice agent Hermes Apollo. Asking it to look at the vault and share ideas works instantly. The agent reads the notes and responds out loud.
There are no numbers on accuracy or speed gains. The main result is that all agents now share one memory and continue each other's work.
Tips, limits, and mistakes to avoid
Julian gives several tips and warnings. First, don't save every single chat or message. Some people ask how to add millions of memories. His honest answer: Obsidian isn't the right tool for that, and you don't need millions unless you're a huge company.
Avoidexotic setups like vector databases for most people. That's too technical. A simple daily log is one prompt: copy, paste, done.
Who this memory system is for and why it matters
This system is for business owners and professionals who use more than one AI agent daily. If you tell one agent something on Monday and it's gone on Tuesday, this fixes that. Instead of re-teaching Claude Code your brand voice after Hermes learned it, they all share the same memory.
Julian says there's a window right now where most business owners haven't figured this out. Most people use one agent at a time and re-explain everything every day. That window won't stay open forever.
By setting up a shared Obsidian memory, you make your whole team of agents smarter. One agent learns a lesson and all others get it. You save time and avoid repeating yourself.
It works for a single person or a small team. The setup is simple and works even if you already have an Obsidian vault with old notes.
