Can a VPS host a personal AI agent? Specs and monthly cost
Short answer: yes, for the agent software itself. A "personal agent" that stays online, remembers context and answers you in a chat app needs a machine that does not sleep, and a small VPS is one such machine. Whether it fits depends on one choice: you call a hosted model through an API, or you run the model yourself. This page uses the official docs of three self-hosted agent projects (OpenClaw, Hermes Agent, NanoClaw) and vendor price pages, all checked on 4 Oct 2026. We have not benchmarked any of it, and we have no search-volume data for this topic, so we make no claim about how popular it is.
What OpenClaw's docs say
- OpenClaw's install page lists Node 24.16+ or 26.1+ and macOS, Linux or Windows as requirements (docs.openclaw.ai/install). Its docs also describe running the Gateway on a Linux VPS (docs.openclaw.ai/vps).
- Its DigitalOcean guide uses a Basic Droplet with 1 vCPU, 1 GB RAM and 25 GB SSD, about $6 a month, and recommends adding swap on 1 GB machines (docs.openclaw.ai/install/digitalocean).
- The same guide says to prefer API-based models over local ones, because local model inference does not fit in 1 GB.
So the server needs enough RAM for the agent process and its tools. A model you run yourself needs far more memory than the plans below. Model API usage is billed by the model provider, varies with how much you use it, and is not in any figure on this page.
Other self-hosted agents, from their own pages
We did not test any of these projects. Stars, releases and commit dates change often, and the data our tools returned was a cached copy of unknown age, so we do not print them. Open the repository for current numbers. We found no rename or end-of-support notice for OpenClaw in its docs or repository page.
| Project | Run requirements stated by the project | RAM or disk minimum stated? | Official repository |
|---|---|---|---|
| OpenClaw | Node 24.16+ or 26.1+; macOS, Linux or Windows (docs) | No minimum found; DigitalOcean guide uses 1 GB plus swap | openclaw/openclaw |
| Hermes Agent (Nous Research) | Linux, macOS (Apple Silicon only) or WSL2 install script; Windows native; Docker image. Current installs run on Python 3.14, which the installer provides. Source install needs Git, curl, tar and SHA-256 tools (docs) | Not on the installation page | NousResearch/hermes-agent |
| NanoClaw | macOS or Linux (Windows via WSL2); Node.js 22+ and pnpm 10+; Docker Engine on Linux; Claude Code for setup and skills (README). Agents run in Docker containers. | Not stated in the README | nanocoai/nanoclaw |
What this means for a VPS: all three list Linux as a supported system, so a Linux VPS is a supported place to run them. Only OpenClaw's DigitalOcean guide gives a plan size (1 vCPU, 1 GB, with swap). For Hermes Agent and NanoClaw the pages we read give no RAM or disk minimum, so we do not name a plan size for them. NanoClaw also needs Docker, which uses memory and disk on top of the agent. Start small on a monthly plan, watch memory, and resize if the agent runs out.
Small VPS plans and prices
| Plan | Spec | Price shown | Per year (x12) | Source, checked 4 Oct 2026 |
|---|---|---|---|---|
| DigitalOcean Basic | 1 vCPU, 1 GiB, 25 GiB SSD | $6.00/mo | $72.00 | digitalocean.com/pricing/droplets |
| DigitalOcean Basic | 1 vCPU, 2 GiB, 50 GiB SSD | $12.00/mo | $144.00 | same |
| DigitalOcean Basic | 2 vCPUs, 2 GiB, 60 GiB SSD | $18.00/mo | $216.00 | same |
| DigitalOcean Basic | 2 vCPUs, 4 GiB, 80 GiB SSD | $24.00/mo | $288.00 | same |
| Hostinger KVM 1 | 1 vCPU, 4 GB RAM, 50 GB NVMe | $6.49/mo shown (regular $19.49) | Renewal text: $11.99/mo = $143.88 | hostinger.com/vps-hosting |
| Hostinger KVM 2 | 2 vCPU, 8 GB RAM, 100 GB NVMe | $8.99/mo shown (regular $24.49) | Renewal text: $14.99/mo = $179.88 | same |
| Hostinger KVM 4 | 4 vCPU, 16 GB RAM, 200 GB NVMe | $12.99/mo shown (regular $42.99) | Renewal text: $28.99/mo = $347.88 | same |
DigitalOcean bills per second with a monthly cap, so the monthly figure is the most a full month costs. Hostinger's page shows the promotional price and says "Renews at ... for 2 years. Cancel anytime." The captured page text does not state the billing term of the promotional price, so we do not compute a promo total; check the term at checkout. The yearly column multiplies the monthly figure by 12 and nothing more: no tax, backups, extra IPs or domain.
How to choose
- API model, one user, a few chat channels: the 1 GiB plan matches what the OpenClaw guide uses, with swap. The 2 GiB DigitalOcean plan at $12.00 gives headroom for $6.00 more a month ($72.00 a year).
- More tools or several agents on one box: start from 4 GB. Hostinger KVM 1 lists 4 GB at a shown $6.49/mo, but its stated renewal is $11.99/mo, so compare the renewal figure ($143.88 a year) with DigitalOcean's 2 GiB plan at $144.00 a year.
- A model you run yourself: none of the plans on this page is sized for it. Check the model's own memory requirements first.
Not covered: other agent projects (we list only ones whose official pages we read today), Muse and similar hosted products (no official self-hosting requirements found), Hetzner, Vultr and other providers (not captured on a vendor page today), and security work such as locking down SSH and keeping API keys out of the agent's workspace. For migration and backup planning, see our migration checklist. Per-agent pages: OpenClaw monthly cost, Hermes Agent requirements and cost, OpenClaw vs Hermes Agent. Prices for shared hosting are on the price ledger. How we check prices: editorial policy.