01What a Hermes Obsidian Second Brain Is and How It Works
A hermes obsidian second brain is one shared folder of notes that all of your AI agents read from and write to. It runs on Obsidian, which is free, so Hermes, Claude Code, Codex, Grok and a voice agent can all work from the same memory.
In this episode, Julian Goldie shows the second brain behind his whole business. It holds over 1,000 notes. Every agent he uses reads from that same folder, so none of them start from zero in a new chat.
The key idea is a nightly shift. Starting at 9 o'clock each night, the agents take turns writing down what they did that day:
- Hermes writes at 9:00.
- Codex writes at 9:10.
- OpenClaw writes at 9:20.
- Claude Code writes at 9:30.
So when Julian wakes up, every agent already knows what every other agent did yesterday. He breaks the whole system into 20 upgrades spread across five levels. Level one gives you a working second brain you can set up in an afternoon.
The later levels turn it into a shared brain that updates itself and helps your agents hand work to each other.
02Level One: Set Up a Basic Hermes Obsidian Second Brain
Level one covers the first five upgrades. With just these, you already have a working second brain. Here is how to set it up:
- Create numbered folders, such as an Inbox, Projects, Resources, Memories and Archive. The numbers keep the folders in the same order every time.
- Leave the structure alone once it is set. Every time you rename a folder, your agent starts saving things in the wrong place.
- Add the Inbox rule. Nothing gets saved at the top level of the vault. If an agent does not know where something goes, it drops it in the Inbox. Write this rule into the instructions every agent reads.
- Put a small file called _index in every folder. It lists what is in that folder and what each file is for.
- Add one file called Home at the very top of the vault that links to every section.
- Write one file that describes you. Julian's is called About Julian, and it sits in Resources. It holds the business facts, the numbers and the voice.
Now an agent opens Home, finds the right folder, reads the index and goes straight to the right note. It does not have to search a thousand files to find one. And when an agent writes something for you, it reads your file first, so every agent works from the same facts instead of making them up.
03Level Two: One Memory Protocol for Every Agent
Level two is where the vault becomes a shared brain for many agents. It covers upgrades six to eight.
- Upgrade 6, the memory protocol: one note in the vault called the AI Memory Protocol. It is a rule book that every agent follows, and it lists where everything lives.
- Upgrade 7, point every tool at the protocol: Claude Code, Codex and Hermes each have their own instruction file that they read first. Instead of writing different rules for each tool, every instruction file says the same thing: here is the vault, here is the protocol.
- Upgrade 8, private memory is only a shortcut: many AI tools now have their own built-in memory, but it is locked inside that one tool. In Julian's protocol, the vault is the system of record. If a fact matters, it goes in the vault where every agent can see it.
The payoff is simple. When Julian adds a rule to the protocol, every agent picks it up without him touching five different tools.
He shares a real example. On the 20th of October, his Claude account hit its weekly limit. Apollo, his voice agent, switched over to Codex and kept going.
Because the memory lives in the vault, nothing was lost. A different model picked up the same notes and the same history.
04Level Three: Nightly Summaries and the Append Only Rule
Level three is how the brain stays up to date on its own. It covers upgrades nine to thirteen.
- Upgrade 9, the nightly shift: every evening, each agent runs a scheduled job and writes its summary.
- Upgrade 10, append only: an agent can add its own section to the daily log, but it can never edit or delete another agent's section. Codex cannot override what Claude wrote, and Hermes cannot wipe what Codex wrote.
- Upgrade 11, one template: every summary uses the same headings: worked on, done, learned and decisions. Because every summary has the same shape, any agent can read another agent's notes in seconds.
- Upgrade 12, a decisions log: one note in Memories where agents add every decision that gets made.
- Upgrade 13, save the insight, not the raw data: keep clean notes instead of huge exports.
The append only rule is one of the most important in the whole system. When several agents write to one brain, it stops them from destroying each other's work.
For the decisions log, Julian gives an example. On the 28th of October, a rule was added that every script request goes through his script writing skill first. That went into the log, so every agent now knows it.
For raw data, he uses a tool called Omi that exports his day into the vault. That export has over 1,200 entries of repeated noise, so the useful parts get pulled into two clean notes: the decisions log and a wins file.
05Level Four: A Wiki, Live Goals and Voice Memory
Level four is where the brain starts helping you think. It covers upgrades fourteen to sixteen.
Upgrade 14 is a wiki inside the vault. There is a folder called Wiki with 167 small notes, one for each person, tool, place and idea that matters to the business. Of those, 67 are tools and 51 are people. The idea comes from Andrej Karpathy, who described an AI wiki where the model keeps small linked pages on every topic.
So when an agent needs to know about one tool or one person, it opens one short page. A log file in that folder records what got added and when, so you can always see where a page came from.
Upgrade 15 is live goals. There is one note called Goals. Julian can add, tick off or remove goals by clicking in the Agent OS, and it saves straight into Obsidian. Right now the active goal is launching a new course in December, and two other goals are already ticked off.
Every agent that reads the vault knows what the business is working toward.
Upgrade 16 is voice memory. Apollo, Julian's voice agent, has its own memory note in the vault:
- He can say "remember I want to launch the new course in December" and Apollo writes it down.
- When Apollo builds his daily briefing, it puts those goals first.
So what you say out loud on Monday shapes what your assistant tells you on Wednesday.
06Level Five: Handover Notes, Skills and Honesty Rules
Level five is the advanced part, for when a whole team of agents shares one brain. It covers upgrades seventeen to twenty.
- Upgrade 17, agents hand work to each other: the summary template has an extra heading called For Other Agents. An agent leaves instructions there for whichever agent picks up the work next.
- Upgrade 18, skills that read from the brain: a skill is a set of saved instructions for one job, like writing a script or editing a video. Instead of each skill keeping its own copy of your facts, the skill reads the About Julian note first every time.
- Upgrade 19, honesty rules: agents write down what they did not check, as well as what they did.
- Upgrade 20, no secrets: no API keys, no passwords and no tokens in the vault, ever.
Here is a real handover. Codex made a batch of images for Julian's Hermes skills content. Under For Other Agents, it wrote to reuse the finished batch and not regenerate everything.
So the next agent did not waste an hour making the same thing again.
The skills rule matters too. If every skill kept its own copy of your facts, you would end up with ten copies and nine would go out of date. With one shared note, update it once and every skill is current the next time it runs.
07Old Way vs New Way With a Shared Second Brain
Julian compares how most people use AI today with how this system works.
The old way:
- You open a chat and explain yourself again.
- You paste in your notes and details every time you switch tools.
- You re-explain what you did yesterday.
- Your AI tools do not talk to each other, so you end up being the messenger between them, especially when one task moves from one agent to another.
The new way:
- You open any agent and it reads the vault first.
- It already knows the business, the goals and what every other agent did yesterday.
- At night, the agents write their own summaries and leave handover notes for each other.
- When one model hits a limit, the next one picks up from the same notes.
You stop being the messenger, because the vault does that job. The honesty rule is a big part of why this works. In one real summary, Codex noted that its review checked the images existed and were the right size, but it did not check every single image.
So the next agent, or Julian, knows exactly what has been checked and what has not. A second brain is only useful if you can trust what is inside it.
08Tips, Common Doubts and Who This Is For
Julian answers three doubts people often have about a hermes obsidian second brain.
- "This sounds like a lot of work." You do not build all 20 upgrades on day one. Level one is five upgrades you can do in an afternoon, and the agents do most of the work for you.
- "I'm not technical." Every file is plain text. If you can write a note on your phone, you can write one in Obsidian, and you can ask your agent in plain English to create the folders and index files.
- "A folder of text files sounds fragile." That is why the append only rule exists. Agents add their own sections and never edit anyone else's. You can also back up the folder like anything else important.
Because it is all plain text on your computer, you own every file and nothing is locked inside another company's app.
This setup suits business owners who use more than one AI agent and are tired of starting from zero in every chat. If you want a ready-made version, the AI Profit Boardroom includes the Agent OS with the Obsidian vault, memory protocol and nightly summaries set up, a 30 day roadmap, a prompt library and four weekly coaching calls. Over 3,400 business owners are inside.