VPSPioneer
Blog Status Support Create account

Hosting for AI agents and LLM apps

AI Agent Hosting

Hosting for AI agents and LLM-powered apps. Bring any model via API, or run small models locally with Ollama on CPU. The server is a Debian 13 VPS in France with Docker installed, a separate non-root agent user, Hermes Agent from Nous Research already running as a service, and the ai-agent-setup wizard that connects a model and a Telegram bot in minutes. Claude Code installs from the same wizard; OpenClaw, Codex CLI, Aider, OpenHands and the usual agent frameworks run alongside with full root.

Deploy your agent Plain Cloud VPS instead
APIAny model from a provider, your account
OllamaSmall models locally, on CPU
Always onA service on a server, not a tab on your laptop
No GPUSaid plainly, so you can plan for it
Hermes Agent logo

Preinstalled

Hermes Agentby Nous Research, ready on first login
One step from the wizard
Claude Code
Install in minutes
OpenClaw, Codex CLI, Gemini CLI, Aider, OpenHands, Goose
Frameworks
LangGraph, CrewAI, AutoGen, Pydantic AI
Local models
Ollama and Open WebUI, CPU only

How it works

Three steps to a running agent

1

Order

Choose a size below. The server is built automatically after payment, with Debian 13, Docker, Hermes Agent and the wizard already on it. The login details arrive in your client area and inbox.

2

SSH in and run ai-agent-setup

Log in once as root. A reminder points you to the wizard; run it and answer a few questions. The agent is installed under its own non-root user and starts as a service.

3

Connect a model and Telegram

Quick Setup signs in with a free Nous Portal account, no API key needed. Or paste your own OpenRouter, Anthropic or OpenAI key. Add a Telegram bot token and talk to your agent from your phone.

Models

Your agent, any model.

The agent is the part that runs here. The model can run at a provider, reached over an API with your own account, or on the server itself with Ollama.

Cloud models over an API

Claude from Anthropic, GPT from OpenAI, the full catalogue behind OpenRouter, or the models that come with a Nous Portal account. You choose the provider and the model, you hold the account, and you can switch at any time by running the wizard again. Usage is billed by the provider at their rates; nothing is marked up by us.

  • Nous Portal: free account, no API key, the default of Quick Setup
  • OpenRouter, Anthropic, OpenAI: your own key with Full setup
  • Change model or provider later without reinstalling

Local models with Ollama

Ollama is not preinstalled, but it installs on any Debian server and you have root. With a small quantized model, around 1 to 8 billion parameters, the agent can answer without any data leaving the server; the M and L sizes have the memory for the larger end of that range. Open WebUI gives it a chat interface if you want one.

No GPU. Local models run slower than GPU hosting; best for light tasks, privacy and testing.

Supported agents

Hermes Agent hosting, a Claude Code server, an OpenClaw VPS: one machine

Three tiers, so there is no doubt about what is already there and what you install yourself.

Tier 1 · Preinstalled

Ready on first login

Hermes Agent by Nous Research is installed under the agent user and configured with ai-agent-setup. A Telegram gateway is part of the setup.

hermes setup
hermes model
hermes update

Tier 2 · One-step install

From the setup wizard

Claude Code, Anthropic's command-line coding agent, is optional and installed by ai-agent-setup when you say yes. You sign in with your own Anthropic account or API key; usage is billed by Anthropic, not by us.

Tier 3 · Compatible

Works on our VPS – install in minutes

Any agent that runs on Linux or Docker will run here; you have full root.

Coding agents
OpenAI Codex CLI, Gemini CLI, Aider, OpenHands, Goose
Personal assistants
OpenClaw
Agent frameworks
LangGraph / LangChain, CrewAI, AutoGen, Pydantic AI
Workflow automation
n8n, preinstalled on our Cloud n8n VPS
Local models
Ollama, with Open WebUI as an optional chat UI (CPU only, small models)
All trademarks belong to their respective owners. VPSPioneer is not affiliated with these projects.

What's included

Set up for an agent, open for everything else

Hermes Agent preinstalledThe open-source agent from Nous Research, installed natively and run as a service that survives reboots.
Setup wizardai-agent-setup handles model, provider and Telegram in one pass; run it again whenever you change your mind.
Docker readyDocker Engine and Compose for the tools and services you add next to the agent.
Non-root agent userThe agent runs under its own account, not as root, so a misbehaving tool call cannot take the whole server with it.
Full root accessIt is a normal Debian 13 server. Install Ollama, a database, Claude Code or a second agent.
IPv4 + IPv6One address of each, with reverse DNS set to the hostname you ask for.
Automatic setup after paymentNothing to wait for: the server is built the moment the payment clears.
Daily snapshotA copy of the previous 24 hours you can restore from the client area.

Use cases

What people run on it

Personal assistant on Telegram

An agent that remembers your context, reads what you send it and reports back, available from your phone at any hour.

Coding agent with Claude Code

Install the Claude Code CLI with one command from the wizard, sign in with your own Anthropic account, and give the agent a server to build and run things on.

Automation alongside n8n

Let n8n handle the scheduled and webhook-driven flows and hand the open-ended steps to the agent. Our Cloud n8n VPS is the same server with n8n preinstalled.

Research and monitoring agents

Watch sources, summarise changes, raise a flag when something matters. Work that is boring for a person and ideal for an always-on agent.

Private LLM app backend

A small app or internal tool that calls a model, with its prompts, logs and data on a server you control rather than a shared platform.

Just want the server?

The plain Cloud VPS is the same hardware with a bare Linux install, for when you would rather build the stack yourself.

Plans

Three sizes, same image on each

Per month, incl. VAT, billed monthly or annually with two months free. Model usage is billed by your provider. Covered by the 30-day money-back guarantee.

AI Agent VPS S

Recommended for Hermes Agent

One agent, always on

vCPU
2 vCores fair-share
Memory
4 GB reserved
Storage
40 GB NVMe
Port
1 Gbps
IP
1 IPv4 + 1 IPv6
Agent
Hermes Agent as a service
GPU
None

Hermes Agent with its memory and a few tools, talking to a model over an API. The size most people need.

$8.49 /mo inc. VAT
$84.90/year
Buy now

AI Agent VPS M

Agent plus its own services

vCPU
4 vCores fair-share
Memory
8 GB reserved
Storage
75 GB NVMe
Port
1 Gbps
IP
1 IPv4 + 1 IPv6
Agent
Hermes Agent as a service
GPU
None

Room for a database, a browser tool or a second agent in Docker alongside Hermes Agent.

$15.99 /mo inc. VAT
$159.90/year
Buy now

AI Agent VPS L

Several agents or heavier tools

vCPU
6 vCores fair-share
Memory
12 GB reserved
Storage
100 GB NVMe
Port
1 Gbps
IP
1 IPv4 + 1 IPv6
Agent
Hermes Agent as a service
GPU
None

Multiple agents, code execution sandboxes or builds, with memory to spare. Still no GPU.

$22.99 /mo inc. VAT
$229.90/year
Buy now
30-day money-back guarantee. No minimum term: cancel any time. Refund policy

Why VPSPioneer

A small host that runs its own servers

Our own infrastructure in a France datacenter
UK company, VPSPIONEER LTD
Monthly billing, or annual with two months free
30-day money-back on every VPS plan
Human support by ticket, from the people who run the platform

Compared

Where a CPU agent server fits

Three ways to keep an agent running, and what each one is good at. No numbers, because yours depend on the model and the hours you keep it on.

 AI agent hosting (this)GPU LLM hostingYour own laptop
CostA fixed monthly VPS fee from $8.49, plus what your model provider charges for usageHourly GPU pricing; the bill grows with every hour the GPU is allocated, used or notNo hosting fee; your electricity and the machine's wear
Always onYes, as a service that restarts on rebootYes, while you keep paying for the instanceOnly while the lid is open and the machine is awake
GPUNone. Models run via API, or small ones on CPU with OllamaYes, for running large models yourselfDepends on the machine; most laptops run only small models
PrivacyAgent memory and files on your server; cloud model calls go to the provider you chose; local Ollama models stay on the serverEverything on the GPU host; depends on the operator and the regionEverything stays with you, as long as the machine does
Setup timeMinutes: the image is preinstalled, the wizard does the restHours: drivers, runtime, model weights, serving stackHours, and again on the next machine
Best forAgents, assistants and LLM apps that call models over an API, plus light local modelsServing your own large model to many users, fine-tuningTrying things out before you commit

FAQ

Questions people ask first

Do you provide GPUs?
No. These are CPU servers. An agent spends most of its time waiting on a model, calling tools and keeping memory, which does not need a GPU. The model itself runs at a provider you reach over an API (Nous Portal, OpenRouter, Anthropic, OpenAI), or, for light tasks, locally with Ollama using a small quantized model on the CPU.
Which models can I use?
Any model your provider offers through its API: Claude models from Anthropic, GPT models from OpenAI, the catalogue behind OpenRouter, or the models available through a Nous Portal account. Locally, Ollama runs small open models in the 1 to 8 billion parameter range at CPU speed.
Do I need an API key?
Not to start. Quick Setup in the ai-agent-setup wizard signs in with a free Nous Portal account and needs no key. If you prefer your own provider, Full setup takes an OpenRouter, Anthropic or OpenAI key, and you pay that provider directly for usage.
Can I run Ollama?
Yes. You have root, so Ollama installs like on any Debian server. Expect CPU speed: a small quantized model answers in seconds, not instantly, and a larger model will be slow and need the bigger plan for memory. Good for private or offline tasks and for testing, not for a chat product with many users.
Is my data private?
The agent, its memory and its configuration live on your own server in our France datacenter, and only you have the login. Whatever the agent sends to a cloud model goes to that provider under its terms; with a local Ollama model nothing leaves the server.
Which AI agents can I run?
Hermes Agent is preinstalled and configured by the wizard. Claude Code is one optional step in the same wizard. Beyond that, anything that runs on Linux or Docker: OpenClaw, OpenAI Codex CLI, Gemini CLI, Aider, OpenHands, Goose, and frameworks such as LangGraph, CrewAI, AutoGen and Pydantic AI. You install and maintain those yourself; the server gives you root and Docker.
Is Claude Code included?
It is an optional install, not part of the image. Say yes in ai-agent-setup and the Claude Code CLI is installed; you then sign in with your own Anthropic account or API key. Anthropic bills the usage, not us.
Can I install other agents or frameworks?
Yes. Full root access and Docker are yours. OpenClaw, LangChain or LlamaIndex apps, a vector database, n8n next to the agent: anything that runs on Linux. We set up Hermes Agent for you; everything else you install and maintain yourself.
What is the refund policy?
VPS plans are covered by our 30-day money-back guarantee: cancel within 30 days of your first order and the payment is refunded in full. After that there is no minimum term and you can cancel at any time. The refund policy on our legal pages has the details.

Deploy your agent today

Pick a size, run the wizard, message your agent from Telegram. From $8.49 a month, 30-day money-back guarantee, cancel any time.

See the plans