Hermes memory setup is simpler than almost everyone expects, because there is nothing to install: Hermes ships with file-based memory already working, so the real setup job is seeding it with your business, correcting it quickly when it gets things wrong, and checking the next day that it genuinely remembered. That is the whole task. The people who get standout results in week one are not doing anything technical — they are feeding the memory real signal instead of leaving the agent to guess.
📺 Watch: Hermes + Jarvis + Obsidian Memory is INSANE!
🔥 Get the Agent OS as a free bonus: AI Profit Boardroom members get the full Agent OS zip, prompt libraries, daily tutorials and weekly live coaching calls. → Get inside
I have been running Hermes across my own business for months now, and a seeded memory versus an empty one is the difference between an agent that sounds like a stranger and one that sounds like a member of staff who has been with you for years. This page is the practical starter — exactly what to do from day one. If you want the deeper explanation of how it all works underneath, my Hermes memory system overview covers the mechanics.
What Hermes Memory Looks Like Before You Touch Anything
Out of the box, Hermes already keeps its memory as plain markdown files on your machine, readable in any text editor. Three pieces do most of the work:
- soul.md — the identity file: who the agent is, who you are, and how it should carry itself.
- Daily notes — a running log of what happened and what was learned in each day's sessions.
- Long-term memory files — the durable facts about you and your business that persist across sessions.
Two behaviours turn those files into a genuine memory system rather than a filing cabinet. First, the agent reads the relevant memory into every session on its own. Second, it writes new learnings back automatically as you work — you never issue a save instruction or manage anything by hand. If you ever want the precise file locations on your machine, the current Hermes docs give the exact steps, but for setup you barely need them: the shaping happens through conversation, not file editing.
Hermes also sits at the centre of my wider Agent OS stack, and memory is the layer that makes everything else in that stack feel personal rather than generic.
The Week-One Hermes Memory Setup Routine
Setup is not an afternoon of configuration. It is a short daily habit for the first week, and it looks like this.
Day one: tell it about your business in plain English
The fastest win available is a simple conversation. Tell the agent who you serve, what you sell, how you price it and how you like things done — in ordinary sentences, exactly as you would brief a new hire on their first morning. No formal documents, no special format, no clever structure.
Cover the essentials: your audience, your offers, your tone of voice, the tools you rely on, the phrases you hate, the standards you hold work to. Every one of those sentences lands in memory and starts paying rent immediately, because the files fill up with signal instead of guesses.
Want my exact day-one seeding prompts? Inside my free community I share the word-for-word briefing I give a fresh Hermes setup, plus the week-one checklist this page is built on. Grab the Hermes seeding routine inside AI Profit Lab and set yours up alongside people doing the same thing this week.
Correct it fast when it gets you wrong
Early on the agent will get things wrong — an old price, a retired offer, a tone you would never use. Correcting it on the spot is not a chore; it is the second half of setup, because corrections persist. Tell it once that you stopped selling something last year and you should not need to say it again. I treat every wrong answer in week one as a free setup prompt: each correction is one more fact the memory holds for good.
Test the recall the next day
Verification is gloriously simple: the next day, ask the agent what it knows about you and your business. What comes back tells you two things — that memory is genuinely writing and reading, and exactly where the gaps are. Whatever it missed becomes that day's seeding conversation. Run that loop for a week and memory stops being a promise on a feature list and becomes something you can watch working.
Shape soul.md Early — It Sets the Tone for Everything
Of the default memory files, soul.md deserves deliberate attention in week one, because it shapes how the agent behaves everywhere, not just what it knows. This is where identity lives: how it should address you, how direct to be, what standards to hold your work to. A few honest sentences about how you like to operate go a surprisingly long way. I have a full guide to shaping soul.md that covers exactly what I put in mine.
The Vault Upgrade Is Step Two of the Memory Build
Once the basics demonstrably work — it knows your business, corrections stick, the next-day recall test passes — the upgrade path is wiring memory into an Obsidian vault, the build I call the Memory Galaxy. It turns those markdown files into a linked, visual second brain you can actually browse. It is brilliant, and it is deliberately step two: build it after the basics are proven, not instead of proving them. The full walkthrough is in my best memory for Hermes guide.
📺 Watch: This Claude + Hermes Setup is INSANE!
Set the Backup Habit While the Memory Is Young
Everything you seed lives in files, and those files travel: a Hermes export carries your memory, your skills and your configuration in one bundle. That makes every hour of seeding portable to a new machine and recoverable after a disaster — but only if exports actually happen. Set the habit in week one while the routine is still forming. I run a weekly export, plus an extra one after any heavy seeding session, and my Hermes backup guide covers the full routine.
Bot Mode: Memory Setup Notes for Multiple Bots
If you plan to run bots, one detail matters at setup time: in bot mode, each bot keeps its own memory. That is a feature — your support bot and your content bot should not share a brain — but it means seeding happens per bot, not once for the lot. Name each bot and give it a clear role early, then brief each one for its own job rather than assuming your main agent's memory carries across. My bot mode guide goes deeper on the whole arrangement.
📺 Watch: Claude + Hermes Setup is WILD
Mistakes I See in Almost Every Hermes Memory Setup
- Dumping everything at once. Pasting your entire business history into day one buries the signal in noise. Seed through short daily conversations across the first week instead — the routine above exists for a reason.
- Never correcting. Some people sigh at a wrong answer and move on. Every uncorrected error is a persistent one; corrections are the setup mechanism, not an interruption to it.
- Treating memory as storage. Memory is not a document archive — it is working context the agent reads every session. Feed it what you want reflected in its answers, not everything you happen to possess. When it eventually needs pruning and tidying, that is a memory management job for later, and it is ten times easier if you seeded with intent.
- Skipping the first export. Seeding for a week without a single backup means your investment lives on one machine. One export removes that risk entirely.
For the habits that sit around memory — briefing style, session hygiene, when to trust it — my Hermes best practices guide rounds out the picture.
Setup Steps at a Glance
| Setup step | Why it matters |
|---|---|
| Tell it your business in plain English | The memory files fill with signal instead of guesses |
| Correct wrong answers on the spot | Corrections persist, so every fix compounds |
| Ask it the next day what it knows | Proves recall is working and exposes the gaps |
| Shape soul.md deliberately | Sets identity and standards for every session |
| Run your first export in week one | Memory, skills and config become portable and safe |
| Name bots and give them roles | Each bot builds its own clean, relevant memory |
Hermes Memory Setup FAQ
Do I need to install anything to get memory working?
No. File-based memory ships working from the first session — the identity file, daily notes and long-term files are all plain markdown on your machine. Setup means seeding and shaping what is already there, not adding components.
What should I tell Hermes first?
Who you serve, what you sell and how you like things done — in plain English, the way you would brief a new team member. Those three areas move the needle faster than anything else you could feed it.
How do I know the memory is actually working?
Ask the agent the next day what it knows about you. Accurate answers prove the write-and-read loop; the gaps show you what to seed next. It is the only test you need in week one.
What if it remembers something wrong?
Correct it in the moment and the correction persists. That is by design — fixing errors is part of setup, not a sign that something is broken.
Will my memory move with me to a new machine?
Yes. A Hermes export carries memory, skills and configuration together, so a move becomes a restore rather than a rebuild. That is exactly why the export habit belongs in week one.
My Verdict: Brief It Like a New Hire, Test It Like One Too
The pattern my Goldie Bench testing keeps confirming is that a well-seeded memory is what separates generic answers from ones that sound like your business — the setup week matters more than any clever tweak afterwards. Talk to it about your business, correct it fast, test the recall, shape the identity file, run the export. Do that for seven days and you will have an agent that knows you; skip it and you will have a clever stranger.
Want the whole routine done with you? Inside my free community I walk through the seeding prompts, the soul.md structure and the recall tests, with people building alongside you. Get my week-one Hermes memory setup checklist inside AI Profit Lab and have yours sounding like your business by this time next week.
