Table of Contents
- Hetzner Hermes agent pricing in September 2026
- Which Hetzner plan does a Hermes agent actually need?
- Step-by-step: Hermes agent on a Hetzner VPS
- Lock it down before you connect channels
- The honest monthly math
- Hetzner gotchas worth knowing before you order
- DIY Hetzner vs managed: when each wins
- Hetzner Hermes agent FAQ
- Can I still get the 5.99 EUR CX23 for a Hermes agent?
- Does Hermes agent run on Hetzner's ARM servers?
- Do I need a GPU for a Hermes agent on Hetzner?
- How much RAM does a Hermes agent need?
- Is Hetzner still cheaper than managed Hermes hosting?
Do not index
Setting up a Hetzner Hermes agent used to be the community's default budget answer: a cheap CX box, thirty minutes of terminal work, done. September 2026 looks different. Hetzner's entire Cost-Optimized line (the famous CX plans and the ARM CAX plans) is marked "currently not available", so the cheapest 4 GB server you can actually order in Europe is the CPX22 at 19.99 EUR a month, and US buyers pay about 60 percent more for their cheapest 4 GB plan. This guide covers what a Hermes agent really costs on Hetzner today, the exact setup commands, the security steps most guides skip, and when a managed instance (Agent37 hosts one from $3.99/mo) is the smarter buy.
Hetzner Hermes agent pricing in September 2026
Hetzner raised cloud prices twice in 2026, with new setup fees on the dedicated-server side in between. The bigger round landed on June 15: the CPX line jumped 2.4x to 3.1x for new orders, while the budget CX and CAX lines rose a gentler 31 to 38 percent. Then the budget lines quietly stopped being orderable. As of this writing, every Cost-Optimized plan shows "currently not available" on hetzner.com, with no restock date.
Here is the real menu today (monthly, including the 0.50 EUR IPv4 add-on Hetzner folds into its displayed prices):
Plan | Specs | Price | Can you order it? |
CX23 | 2 vCPU, 4 GB RAM, 40 GB NVMe | 5.99 EUR | No, out of stock |
CAX11 (ARM) | 2 vCPU, 4 GB RAM, 40 GB NVMe | 6.49 EUR | No, out of stock |
CPX12 | 1 vCPU, 2 GB RAM, 40 GB NVMe | 11.99 EUR | Yes (EU) |
CPX22 | 2 vCPU, 4 GB RAM, 80 GB NVMe | 19.99 EUR | Yes (EU), cheapest 4 GB |
CPX11 (US) | 2 vCPU, 2 GB RAM, 40 GB NVMe | $21.09 | Yes (Ashburn/Hillsboro) |
CPX21 (US) | 3 vCPU, 4 GB RAM, 80 GB NVMe | $38.09 | Yes, cheapest 4 GB in the US |
Three billing details worth knowing. EU servers include 20 TB of traffic while US plans include 1 to 2 TB. Billing is hourly, rounded up, and capped at the monthly price. And Hetzner bills a server until you delete it, not until you power it off.
If those numbers surprise you, you are not alone: most guides ranking for this topic still quote the pre-June CX22/CX23 at 4.50 to 5 EUR. Our best VPS for Hermes agent comparison has the wider seven-provider picture, and our cheap Hermes hosting breakdown tracks how the budget tier got here.
Which Hetzner plan does a Hermes agent actually need?
Hermes agent is lean. The official Docker docs put the floor at 1 GB RAM, 1 core, and 500 MB of disk, with 2 to 4 GB recommended, and at least 2 GB whenever browser automation (Playwright plus Chromium) is enabled. No GPU is needed on the server: the model runs remotely through whatever API you configure, so a shared-vCPU plan is exactly the right shape.
Practically, that maps to two choices today:
- CPX12 (2 GB, 11.99 EUR) works for a headless agent. Install with the
--skip-browserflag and skip Playwright.
- CPX22 (4 GB, 19.99 EUR) is the safe pick if you want browser tools, multiple channels, and room for skills and session history to grow.
One fact none of the ranking guides mention: the official
nousresearch/hermes-agent Docker image is published multi-arch, linux/amd64 and linux/arm64. Hermes runs fine on Hetzner's ARM CAX plans, which were the best price-performance deal on the menu. Worth remembering if the Cost-Optimized line comes back in stock.Step-by-step: Hermes agent on a Hetzner VPS
The whole install takes about 20 minutes on a fresh Ubuntu 24.04 server.
1. Create the server. In the Hetzner Cloud console pick Ubuntu 24.04, a CPX22 (or CPX12 for headless), and add your SSH key. Then connect over real SSH from your terminal. Do not paste commands through Hetzner's browser-based console: the official Hermes docs warn it corrupts special characters (a
: can arrive as ;), which turns config lines into mystery bugs.2. Create a non-root user. The Hermes security docs are explicit: never run the gateway as root.
adduser hermes
usermod -aG sudo hermes
loginctl enable-linger hermes
su - hermesThe
enable-linger line matters: it lets the hermes user's background service start at boot and survive logouts.3. Run the official installer.
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bashThis is the only supported Linux path (the hermes-agent packages floating around PyPI and npm are explicitly unsupported by the project). It brings in Python 3.11 via uv, Node.js v26, ripgrep, and ffmpeg. Existing system Node 22.22+ or 24.11+ is used as-is. On a 2 GB box, pass the skip flag:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash -s -- --skip-browser.4. Connect a model. Run
hermes setup --portal for the one-OAuth Nous Portal route, or bring your own key from OpenRouter, Anthropic, OpenAI, Google, or a local endpoint. Hermes itself is free, MIT-licensed software; the model API is what you pay for.5. Set up channels and make it permanent.
hermes gateway setup # wizard: Telegram, WhatsApp, Discord, Slack, 25+ platforms
hermes gateway install # installs the systemd service
hermes gateway status
hermes doctor # verifies the whole installThe gateway is a single background process that holds all your platform connections and runs the cron scheduler. That persistent process is the entire reason a Hermes agent needs an always-on box instead of your laptop.
If you would rather run everything in a container, the same setup works with the official image; our Hermes Docker guide walks that path end to end.
Lock it down before you connect channels
Hermes ships with sane defaults, and you should keep them.
- Messaging is default-deny. Unknown DMs get an 8-character pairing code you approve with
hermes pairing approve. Keep it that way; settingGATEWAY_ALLOW_ALL_USERS=trueon a public bot is how strangers end up with shell access to your server.
- Leave the dashboard on loopback. The web dashboard binds to 127.0.0.1:9119 by default, and since the mid-2026 hardening a public bind forces a login instead of silently exposing the UI. Reach it over an SSH tunnel (
ssh -L 9119:localhost:9119 root@your-server) or Tailscale instead of opening the port.
- Contain the shell. Set
terminal.backend: dockerin config.yaml so commands the agent runs execute inside a hardened container, and keep secrets in~/.hermes/.envwithchmod 600.
The honest monthly math
The agent is free. You pay for a server and for model tokens, and the tokens are usually the bigger line item.
Route | Monthly cost | Plus |
Hetzner CPX22 (EU, 4 GB) | 19.99 EUR (~$23.59) | model API |
Hetzner CPX12 (EU, 2 GB, headless) | 11.99 EUR | model API |
Hetzner CPX21 (US, 4 GB) | $38.09 | model API |
CX23/CAX11, if restocked | ~6 EUR | model API |
Nous Hermes Cloud (Medium, 24/7) | ~$16.80 | inference and tools |
Agent37 managed Basic | $3.99 | your model key |
Agent37 Cloud dedicated instance | $4.94 | usage you meter |
Two notes on the non-Hetzner rows. Nous's own Hermes Cloud is still labeled preview: $0.56/day for the Medium instance (2 GB RAM, 4 vCPU), $10 minimum in credits or an active Portal subscription to deploy, and its channel list stops at Telegram, Discord, Slack, email, and CLI, so WhatsApp and Signal stay self-hosted advantages. Agent37 runs managed Hermes instances with 4 GB RAM on the $3.99 Basic tier, which is less than any server Hetzner will sell you today.
Model spend varies more than hosting: light use on efficient models runs a few dollars a month, heavy frontier-model use can pass $20. Budget for it either way.
Hetzner gotchas worth knowing before you order
Account verification can stop you at the door. Hetzner's fraud screening may ask new customers for a government ID or an advance card payment, and some registrations are rejected with no stated reason. Keep your first order small; a big first order is exactly what gets flagged.
The US price trap. Among shared-vCPU plans, Ashburn and Hillsboro sell only the older CPX x1 line (dedicated-vCPU CCX plans exist there too, at much higher prices), at roughly 1.5 to 1.8 times EU pricing with a fraction of the traffic allowance. If you are US-based and latency matters, compare against US-native providers before defaulting to Hetzner.
Stock is a moving target. The Cost-Optimized line may return; Hetzner frames the scarcity as limited batches of proven hardware. Check the pricing page the day you order, not a blog post from spring.
Deleting, not stopping, ends billing. A powered-off server bills at full rate until destroyed.
DIY Hetzner vs managed: when each wins
Pick the Hetzner route if you are in Europe, comfortable in a terminal, and want the full self-hosted surface: every one of Hermes's 25+ channels including WhatsApp and iMessage bridges, root on the box, and 80 GB of NVMe you can use for anything else. At 19.99 EUR it is no longer the bargain the older guides promise, but it is a solid machine from a serious provider, and if the CX/CAX plans restock the value story comes right back.
Pick managed if the sysadmin role is the part you do not want. Agent37 runs your Hermes agent always-on from $3.99/mo: one-click deploy, a task board, full web terminal, file browser, and 1,000+ app integrations, with your API key going straight from the container to the model provider. Builders who want an API instead of a dashboard can provision the same thing programmatically on Agent37 Cloud, where a dedicated 2 vCPU / 4 GB instance with 6 GB of disk is $4.94/mo and new workspaces get a $1 starter credit with no card. Honest caveat: at these prices Basic comes with community support, not a human on call; email and chat support start on the Pro tier.
Hetzner Hermes agent FAQ
Can I still get the 5.99 EUR CX23 for a Hermes agent?
Not right now. As of September 2026 the whole Cost-Optimized line, CX and ARM CAX both, is marked unavailable on hetzner.com with no restock date. The cheapest orderable 4 GB plan is the CPX22 at 19.99 EUR/mo.
Does Hermes agent run on Hetzner's ARM servers?
Yes. The official Docker image ships linux/arm64 alongside amd64, and the installer supports aarch64 Linux. The catch is availability: the CAX plans are out of stock along with the CX line.
Do I need a GPU for a Hermes agent on Hetzner?
No. The model runs through an API (Nous Portal, OpenRouter, Anthropic, OpenAI, and others), so the server only runs the gateway process. A shared-vCPU plan is the right fit; GPU servers are wasted money here.
How much RAM does a Hermes agent need?
The official floor is 1 GB with 2 to 4 GB recommended, and at least 2 GB if browser automation is on. On Hetzner today that means CPX12 for a headless agent or CPX22 to run everything comfortably.
Is Hetzner still cheaper than managed Hermes hosting?
Not at current stock. A 4 GB Hetzner box is 19.99 EUR (EU) or $38.09 (US) per month, while managed Hermes instances at Agent37 start at $3.99/mo with 4 GB RAM. If the CX/CAX lines restock near 6 EUR the gap shrinks to about three dollars a month, and you trade your time for what remains.
Ready to skip the server admin? Deploy an always-on Hermes agent on Agent37, from $3.99/mo, live in one click.
