agentos.guide › blog
The best Hermes Agent VPS for most people is the Hostinger KVM 2, with 2 vCPU and 8GB of RAM at $8.99 a month on promo and a one-click Hermes template.
Hermes Cloud from Nous Research is my second pick if you want no server at all, and Oracle Cloud's Always Free tier is the best free option.
Below you'll get the ranked list, the size you actually need, a quick-start for an Oracle VPS and how an always-on server fits into an Agent OS.
Here is the whole answer before anything else.
| Rank | Option | Specs I'd choose | Price seen on 10 Oct 2026 | Best for |
|---|---|---|---|---|
| 1 | Hostinger KVM 2 | 2 vCPU, 8GB RAM, 100GB NVMe | $8.99 a month promo, $14.99 renewal | Most people who want a real server |
| 2 | Hermes Cloud | Standard: 2 vCPU, 2GB RAM | $0.51 a day running, $0.03 stopped | People who want no server at all |
| 3 | Oracle Cloud Always Free | 2 OCPU, 12GB RAM on Arm | Free | A capable server on a zero budget |
| 4 | Hetzner Cloud | CPX22: 2 vCPU, 4GB RAM, 80GB NVMe | Check current pricing | A server in a European data centre |
| 5 | DigitalOcean | 2 vCPU, 4GB RAM, 80GB SSD | $24 a month, or $12 for 2GB | First-time server owners |
| 6 | Contabo | Cloud VPS 4: 4 vCPU, 8GB RAM | From €4.40 a month on 24 months | The most RAM for the money |
| 7 | Vultr | Check the plan list | Check current pricing | People who already use Vultr |
I checked each price on the provider's own page on 10 October 2026.
I run Hermes every day on my Mac and I've deployed Hermes Cloud, but I haven't personally run Hermes on every VPS in that table.
The VPS ranking is therefore based on published specs, published prices and the official Hermes docs.
Hetzner's prices didn't load for me and Vultr's pricing page blocked my check, so I've told you to check those yourself instead of guessing.
I used four tests, and you can repeat every one of them.
The KVM 2 plan gives you 2 vCPU, 8GB of RAM, 100GB of NVMe storage and 8TB of bandwidth.
The promo price is paid upfront, and the page listed a renewal price of $14.99 a month.
Hostinger's own Hermes Agent cost guide says its template installs Hermes Agent in one click.
That same guide recommends at least 2 vCPU and 8GB of RAM for a stable Docker deployment.
Hermes Cloud is the hosted version of Hermes Agent, run by Nous Research.
Standard has 2GB of RAM, 2 vCPUs and 10 concurrent sessions, and Pro has 4GB of RAM, 4 vCPUs and 20 concurrent sessions for $0.99 a day running.
You need a $2 minimum credit balance or an active Nous subscription to deploy, and inference and tool usage are billed separately.
I covered the whole deploy flow in my Hermes Cloud guide.
Oracle describes its free Arm allowance as equivalent to 2 OCPUs and 12GB of memory for Always Free tenancies.
You also get 200GB of block storage and 10TB of outbound data a month.
Hermes lists Linux on aarch64 as a Tier 1 platform, so the Arm shape is supported.
The honest caveat is that Oracle can reclaim Always Free instances it judges to be idle over a 7-day period.
Hetzner's CPX22 lists 2 vCPU, 4GB of RAM and 80GB of NVMe, with data centres in Germany and Finland plus US and Singapore locations.
DigitalOcean's 4GB Droplet with 2 vCPUs costs $24 a month, and its 2GB Droplet costs $12 a month.
Contabo's Cloud VPS 4 starts from €4.40 a month, and Contabo states that this is the effective rate on a 24-month subscription.
I couldn't see Vultr's plans today, so treat it as an option to price up yourself.
🔥 Want the exact Hermes setup I run inside my Agent OS?
Inside the AI Profit Boardroom you get the installable Agent OS, step-by-step video tutorials, four coaching calls a week and 3,400+ members building real automations.
The answer depends on where the model runs, because Hermes is the harness and the model is the brain.
| Where the model runs | Server size I'd choose | Why |
|---|---|---|
| An API, with chat and scheduled jobs only | 2GB RAM and 1 to 2 vCPU | The server only runs Hermes and the gateway. |
| An API, with browser tools switched on | 4GB to 8GB RAM and 2 vCPU | The Chromium browser tool needs memory on top of Hermes. |
| A local model on the server itself | A GPU server, not a normal VPS | A CPU-only VPS runs agent-grade local models far too slowly. |
| A local model on your own Mac | No VPS needed | Your Mac does the thinking and costs nothing extra. |
The official installation docs don't publish a minimum RAM or CPU figure.
The Hermes README says you can run it on a $5 VPS, and that is true for a light agent.
The installer adds a browser tool with a pinned Chromium by default, which is why I'd avoid 1GB plans.
Nous sizes its own smallest Hermes Cloud instance at 2GB of RAM and 2 vCPUs, so I treat 2GB as the practical floor.
If picking the brain is your real question, my best Hermes Agent model ranking covers it.
One of my viewers asked how to set Hermes up on an Oracle VPS, so here are the real steps.
The same commands work on any Ubuntu server from the third step onwards.
Create a compute instance in the Oracle Cloud console with a Canonical Ubuntu image.
Choose the VM.Standard.A1.Flex shape and set it to 2 OCPUs and 12GB of memory.
Save the SSH key the console offers, then connect from your own computer.
ssh -i ~/.ssh/your-oracle-key.key ubuntu@YOUR_SERVER_IP
sudo apt update && sudo apt upgrade -y
sudo adduser --disabled-password --gecos "" hermes
sudo -iu hermes
The Hermes docs say to run the installer as the intended service user, so you install under this new user.
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
source ~/.bashrc
The installer puts Hermes in your home folder and sets up its own Python and Node.js.
hermes setup
hermes doctor
The wizard configures your model and tools, and the doctor command tells you what is missing.
You can run hermes setup --portal to log in to Nous Portal and switch on the Tool Gateway in one go.
Message @BotFather in Telegram, send /newbot and copy the token.
Message @userinfobot to get your numeric user ID.
hermes gateway setup
Set TELEGRAM_ALLOWED_USERS to your own user ID when the wizard asks.
Telegram uses polling by default, so the server calls out to Telegram and you don't open any inbound port.
hermes gateway install
hermes gateway start
Switch back to the ubuntu user and enable lingering so the service survives a logout.
sudo loginctl enable-linger hermes
The docs also describe sudo hermes gateway install --system for headless hosts that should come back at boot.
Send your bot a message, and your agent is now running around the clock.
An agent with terminal access on a public server deserves real care.
GATEWAY_ALLOW_ALL_USERS=true.~/.hermes/.env on the server and never paste them into a chat.hermes update regularly and hermes security to audit plugins and MCP servers.hermes backup before big changes, because hermes import can restore that zip on a new server.🔥 Want the exact Hermes setup I run inside my Agent OS?
Inside the AI Profit Boardroom you get the installable Agent OS, step-by-step video tutorials, four coaching calls a week and 3,400+ members building real automations.
This is where a server stops being a tech project and starts being useful.
In my Agent OS, each machine has one clear job.
| Machine | Its job in my Agent OS | What runs there |
|---|---|---|
| My Mac | The workshop where things get built | The Agent OS dashboard, local models, builds and screen recordings. |
| A VPS or Hermes Cloud | The night shift that never sleeps | The Hermes gateway, Telegram and scheduled jobs. |
| My phone | The remote control | Telegram messages to the agent, plus the dashboard over Tailscale. |
My Mac is where I build, because that's where the dashboard, the local models and the galleries live.
A server is where I'd leave the jobs that must run while the Mac is shut.
The gateway on the server answers Telegram and runs scheduled jobs, such as a morning briefing or a weekly report.
My phone is how I talk to the agent when I'm away from the desk.
I don't open the dashboard to the public internet.
Tailscale puts your server, laptop and phone on one private network with no open ports.
curl -fsSL https://tailscale.com/install.sh | sh
sudo tailscale up
My Agent OS on your phone guide shows the same Tailscale idea end to end.
You don't have to start from scratch if Hermes already runs on your Mac.
The hermes backup command zips your Hermes home, and hermes import restores it on the new machine.
Check your keys and channel settings after the import, because a server should only hold the keys it needs.
| Route | What it costs | What you manage | When I'd choose it |
|---|---|---|---|
| Your own VPS | A fixed server bill plus model usage | Updates, security and backups | You want control and a predictable bill. |
| Local Mac | No hosting bill | Keeping the machine awake | You build at a desk and want local models. |
| Hermes Cloud | $0.51 to $0.99 a day running, plus model usage | Almost nothing | You want an always-on agent today. |
Most people I talk to don't need a server on day one.
Start on your own machine, and add a server when a job has to run while you sleep.
My best Hermes Agent setup ranking compares the full stacks.
| Cost line | What the provider's page showed on 10 Oct 2026 |
|---|---|
| Nous Portal Free | The plan is $0 and includes free models only. |
| Nous Portal Plus | The plan is $20 a month and includes $22 of monthly credits. |
| Hermes Cloud Standard | The instance is $0.51 a day running, which is about $15.30 for 30 days non-stop. |
| Hostinger KVM 2 | The server is $8.99 a month on promo and $14.99 a month on renewal. |
| Oracle Always Free | The server is free within the Always Free limits. |
The server is the predictable part of the bill, and the model is the part that grows with use.
Free models keep the model line at zero, but they change over time and get rate limited at busy hours.
My run Hermes free forever guide covers the free brains I've tested.
There are also third-party managed hosts, and I looked at one in my earlier best Hermes VPS post.
My pick is the Hostinger KVM 2, with 2 vCPU and 8GB of RAM at $8.99 a month on promo and $14.99 on renewal when I checked on 10 October 2026. Hermes Cloud is the best no-server option and Oracle Cloud Always Free is the best free option.
The official docs don't publish a minimum and the README says it can run on a $5 VPS. I treat 2GB as the floor, because that is what Nous gives its smallest Hermes Cloud instance, and 4GB to 8GB as comfortable with browser tools.
Create an Ubuntu instance on the Arm A1 Flex shape, connect over SSH, create a non-root user, run the official install script, run hermes setup, connect Telegram with hermes gateway setup, then install the gateway as a service and enable lingering.
No. The Hermes Telegram gateway uses polling by default, which only makes outbound requests, so no inbound port is needed.
Use your Mac to build and to run local models, and use a VPS or Hermes Cloud for jobs that must run while your Mac is shut. Hermes Cloud removes the server admin, while a VPS gives you control and a fixed bill.
Pick the option that matches how much server work you're willing to do, and run the quick-start today.
Give the server one always-on job first, and you'll quickly see whether you've found your best Hermes Agent VPS.
About Julian
I'm Julian Goldie, an AI entrepreneur, SEO expert and the founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
My YouTube channel has 400,000+ subscribers, and I run Goldie Agency, a seven-figure SEO agency.
→ Hermes Cloud: The 60-Second AI Employee
→ Your Agent OS On Your Phone With Tailscale
The installable system behind every guide on this site — dashboard, agents, memory and pipelines — updated as new tools land, with 3,400+ members and four coaching calls a week.
Join the AI Profit Boardroom →📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉