How do I run Hermes Agent 24/7 on a VPS?
Install Hermes on an Ubuntu VPS with its install script, set an OpenAI-compatible provider in config.yaml, and run hermes gateway install with lingering on so the messaging gateway survives logouts and reboots. Hermes needs 1 GB of RAM and recommends 2 to 4 GB, which costs €5.99 a month at Hetzner or $10 to $12 at Vultr and DigitalOcean.
# Ubuntu VPS, as a normal user with sudo
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash -s -- --non-interactive
# Your key, read through key_env in config.yaml
echo 'STANDARD_COMPUTE_API_KEY=YOUR_API_KEY' >> ~/.hermes/.env
chmod 600 ~/.hermes/.env
Keys and values follow the Hermes providers docs and match the profile our launcher writes for Hermes. Run hermes model instead if you prefer a wizard: choose a custom endpoint and enter the same base URL, key and model.
Which VPS fits Hermes
| Server | CPU and RAM | Per month | Note |
|---|---|---|---|
| Oracle Always Free | 2 Arm OCPUs, 12 GB | $0 | Idle instances may be reclaimed |
| Hetzner CX23 | 2 vCPU, 4 GB | €5.99 | EU only; marked unavailable on Oct 13 |
| Vultr Regular | 1 vCPU, 2 GB | $10 | The low end of Hermes' 2 to 4 GB |
| Hostinger KVM 1 | 1 vCPU, 4 GB | $11.99 | Renewal price; $6.49 on a prepaid intro term |
| DigitalOcean Basic | 1 vCPU, 2 GB | $12 | 2 vCPU and 4 GB is $24 |
Prices on October 13, 2026 from Oracle, Hetzner (excluding VAT, with IPv4), Vultr's public plans API, Hostinger and DigitalOcean. Hermes' Docker guide asks for 1 GB, 1 core and 500 MB of disk at minimum, and recommends 2 to 4 GB, 2 cores and 2 GB of disk.
Where Hermes spends model calls when nobody is chatting
| Source | Default | Setting |
|---|---|---|
| Side tasks: compression, titles, vision | The main model | auxiliary.<task>.model |
| Delegated subagents | The main model | delegation.model |
| Cron jobs | An agent turn | cron.model, or --no-agent for jobs that need no model |
| Context compression | At 50% of the window | compression.threshold |
From the Hermes configuration docs, checked October 13, 2026.
When a cron job runs all day
A Hermes gateway answers messages, cron jobs and webhooks with nobody watching, and on a metered key each turn is a billed request.
| Sonnet 5.5, metered | Opus 5.5, metered | Standard Compute | |
|---|---|---|---|
| One busy day | About $35 | About $70 | Up to the $20 budget of the $19 plan |
| Busy every day for a month | $1,048 | $2,095 | $19 |
| What stops it | A spend limit, if you set one | A spend limit, if you set one | The budget: new requests get HTTP 402 |
A busy day is our 90th-percentile agent-day: 775 requests and 79M tokens over a median of 14 active hours, priced at list price with the cache hits it really had (method). A day like that can use the whole Starter budget; smart routing stretches it, and larger plans raise it.
The catch
A Docker install of Hermes can't update itself: you update by pulling a new image, and the ~/.hermes mount is what carries memory, skills and config across. Back it up before each pull.
Where Standard Compute fits
Standard Compute is an OpenAI- and Anthropic-compatible API at https://api.stdcmpt.com/v1, with plans from $19 a month. Each plan has a fixed compute budget ($20 on Starter); when it is used, new requests get HTTP 402 and running ones finish, so there is no overage. A cron job stuck in a loop stops at the plan price, and smart routing sends Hermes' side tasks to efficient models.
FAQ
What is the best VPS for Hermes Agent?
Any Ubuntu server with 2 to 4 GB of RAM. On October 13, 2026 that meant a Hetzner CX23 at €5.99 in Europe, a 2 GB Vultr plan at $10 or a 2 GB DigitalOcean droplet at $12. Oracle's Always Free tier works at $0 if you accept its idle-reclamation rule.
How much RAM does Hermes Agent need?
1 GB at minimum and 2 to 4 GB recommended, with 1 to 2 CPU cores and 500 MB to 2 GB of disk, per Hermes' Docker guide.
How do I keep Hermes running after I log out of the server?
Run hermes gateway install, then sudo loginctl enable-linger $USER. The user service then starts at boot and keeps running with nobody logged in; Hermes' docs prefer this to a system service because it needs no root.
Can I run Hermes Agent in Docker on a VPS?
Yes. The official image is nousresearch/hermes-agent: mount ~/.hermes to /opt/data, run the gateway run command and set restart to unless-stopped. Hermes recommends 2 to 4 GB of memory for the container.
Can Hermes use an OpenAI-compatible API like Standard Compute?
Yes. Add it as a named provider in config.yaml with its base URL, key_env and chat_completions transport, then set model.provider to custom:standardcompute. The hermes model wizard sets up the same thing.