An agent that is supposed to work while you sleep cannot live on a laptop that goes to sleep with you. That is the entire case for a Hermes agent Docker setup: you take Hermes, the free open-source agent from Nous Research, and give it a clean, sealed, restartable home that stays up whether your Mac is open or not. I run Hermes every day across Telegram, WhatsApp and my desktop, and the moment it started doing scheduled work for me, "where does it live" became the most important question in the setup.
📺 Watch: How to Run Hermes Agent for FREE
🔥 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
One honesty note first, because I would rather lose a click than your trust. Hermes moves fast, and exact container commands belong in the official Nous Research GitHub and docs, not in a blog post that could be six months stale. So this page teaches the concepts that make containerised Hermes work — what to persist, what to back up, what Docker actually buys you — and points you at the official instructions for the specifics. Concepts age well. Copy-pasted commands do not.
Why a Hermes agent Docker setup earns its keep
For an always-on agent there are four boring, practical wins, and they are the same four wins containers give any long-running service. This is standard container practice applied to Hermes, not some secret Hermes feature.
- Isolation. The agent and its dependencies live in one container that cannot fight with the rest of your system. No version clashes with whatever else is on the box, no mystery breakage because another tool updated a shared dependency.
- Reproducibility. The same container behaves the same on your laptop and on a VPS. Test the setup at home, then run the identical thing on a server. When something works it keeps working, because the environment is pinned.
- Restarts. Servers reboot. Providers patch things. A container configured to restart comes back up on its own — and for an agent whose whole job is running scheduled work at 3am, coming back without you noticing is the entire feature.
- Portability. Outgrow a cheap VPS? Move the container and its data to a bigger one. You are moving a sealed box plus a folder of state, not re-living your original setup evening from memory.
Hermes is exactly the kind of stateful, long-running, needs-to-survive-reboots service containers were designed for.
The golden rule: the container is disposable, the agent is not
Here is the concept that decides whether this goes well or ends in tears. Containers are built to be thrown away and rebuilt. Your agent is not. Hermes keeps everything that makes it yours in folders on the machine — profiles are isolated folders holding config, memory, skills, credentials and chat history. If those folders exist only inside the container, deleting the container deletes your agent. Months of memory, gone in one rebuild.
The standard Docker answer is volumes: map the state folders to storage outside the container, so the container stays disposable while the agent's brain persists. Rebuild or replace the container as often as you like — memory, skills, profiles and config sit safely on the host and get mounted back in every time. It is the golden rule of every stateful container, and it applies to Hermes unchanged: losing the container should never mean losing the agent.
Get that one decision right and everything else about running Hermes in Docker is detail. Get it wrong and no amount of clever setup will save you.
If you want your agent running around the clock without babysitting servers, the AI Profit Boardroom walks hosting setups like this step by step. → Get the hosting playbook
📺 Watch: Hermes AI Agents Just Went Portable
Export first: the seatbelt for any container change
Volumes protect you from container rebuilds. They do not protect you from fat-fingering the volume setup itself, or from the host dying. My rule before any hosting change: run an export first. The export command packages the whole agent — persona, skills, memory, cron jobs, plugins — into one file with the API keys stripped out. That file is your disaster recovery: whatever happens to container, volume or server, the agent can be rebuilt from it on any machine.
The routine is written up in my Hermes backup guide, and the backup and restore walkthrough covers the round trip. Export before you touch anything, store the file off the server, and every container experiment becomes reversible.
The always-on payoff
Why bother with any of this? Because the best parts of Hermes only shine when it never sleeps. Cron jobs run your morning briefings and reports on schedule. Loops and goals keep recurring work grinding away without you prompting every step. And the gateway — the piece connecting Hermes to Telegram, Discord, Slack, WhatsApp and email — is only as available as the machine it runs on. On a laptop it goes down with the lid; in a container on a VPS it is simply always there.
The split setup is confirmed to work. The gateway lives on the server; the desktop app on your Mac, Windows or Linux machine connects to it remotely — same agent, same memory, same sessions, from phone or desktop. For private access without exposing anything to the public internet, my Tailscale gateway setup is the pattern I use: the server joins a private network only your own devices can see.
📺 Watch: Grok Bot DESTROYS Hermes Agent?
Three honest routes to a hosted Hermes
This is where most articles would paste a container recipe. I am not going to, because Nous Research iterates quickly and a stale command block is worse than none. Three real routes.
Route one: the container route, via the official docs
If you want Hermes in Docker, go to the official Hermes GitHub repository and docs and follow the current container guidance there. The exact commands belong where they get updated. What this page arms you with is the checklist that does not change: where does the state live, is it on a volume, does the container restart on its own, are keys injected at runtime, and did you export first.
Route two: the bare VPS install, the simplest supported path
Honest truth: you do not need Docker for an always-on Hermes. The simplest supported route is a plain VPS — Hostinger is the sort of thing I have pointed people at — with the official one-line terminal installer run straight on the server. Nothing new to learn, identical always-on payoff. My Hermes hosting guide walks that route end to end. If "volume mapping" made your eyes glaze over, this is your path.
Route three: local machine, maximum privacy
The third route is no hosting at all: run Hermes on a machine in your own house. Maximum privacy, zero server cost, and you can pair it with a local model setup so even the thinking happens on your own hardware. Which brain the hosted agent runs is a side-by-side call — exactly what I built Goldie Bench for, my rig for testing models head to head. The trade-off: your "always on" is only as always-on as your home machine.
Who should actually bother?
If you run 24/7 automations, cron routines, loops, or a gateway your team messages all day, a hosted Hermes is worth the evening it takes. That is my situation — my whole Agent OS depends on agents that do not stop when I do.
If you use Hermes casually — chat, the odd task, no scheduled work — the desktop app is enough. Come back the day you wish the agent had done something while you slept.
The limits nobody puts in the sales pitch
- A container is not a security boundary against your own agent. Hermes still acts with whatever permissions and credentials you hand it, container or not. Approvals and sensible tool permissions still matter.
- Small VPSes have small ceilings. Agent plus gateway on a budget box can hit memory and CPU limits. Standard practice: cap the container's resources so a runaway process cannot take the server down.
- Keep secrets out of the image. Basic hygiene: API keys get injected at runtime, never baked into an image that might get pushed or shared. The export file strips keys for the same reason.
The routes at a glance
| Route | Effort | Best for | The catch |
|---|---|---|---|
| Container on a VPS | Medium — follow the official docs | Clean, rebuildable, portable always-on setups | Exact steps live in the official docs; persistence is on you |
| Bare VPS, one-line installer | Low | The simplest supported always-on route | Less disposable environment; rebuilds are manual |
| Local machine at home | Low | Maximum privacy, zero hosting cost | Only as always-on as your house |
Hermes agent Docker: quick answers
Can Hermes actually run in Docker?
Check the official Hermes GitHub and docs for the current container route — that is where live guidance belongs. What does not change is the conceptual side: persist the state folders, export before changes, inject keys at runtime, make the container restart on its own.
Will my agent's memory survive container updates?
Yes, if you follow the golden rule: state folders on volumes outside the container, so rebuilds never touch memory, skills or profiles. Keep an export file on top and even a destroyed volume is recoverable.
Can I run this on a cheap VPS?
Generally, yes. Unless you are running a local model, the heavy thinking happens on the model provider's side, not your box, so a modest VPS copes with the agent and gateway. Watch memory on the smallest tiers.
Do I need Docker at all?
No. The one-line installer straight onto a VPS is the simplest supported route, and a home machine works too. Docker adds isolation, reproducibility and painless rebuilds — lovely to have, required never.
How do I reach a hosted Hermes remotely?
Message it through the gateway on Telegram, WhatsApp and the rest, or connect the desktop app to the remote gateway for the full interface. For private access from your phone, a Tailscale-style private network is my preferred pattern.
My verdict
Running Hermes in Docker is the tidy version of a decision I would make anyway: an agent that works around the clock deserves a home that is isolated, reproducible and self-restarting. Learn the concepts here, take the current commands from the official docs, persist the state folders as if your agent's life depends on them — it does — and export before every change. Do that and the box it lives in becomes the most boring part of your stack, which is exactly what you want.
If you want an agent that works around the clock while you get your evenings back, check out the AI Profit Boardroom — inside you get my Hermes setup and hosting tutorials, the full Agent OS, four weekly coaching calls, daily updates and 3,700+ business owners building the same systems. → Get your agent running 24/7
