AI Model VPS HostingSelf-Hosted Models, Honestly Sized
Hosting your own AI model means your prompts and documents never leave your server. Our VPS plans have no GPU, so this suits small open models and private, low-volume work. We set up the server and runtime; you choose the models.
Full root & SSH NVMe storage Daily backups Anti-DDoS 99.9% uptime SLA 11 locations
-
Data stays on your server
Prompts and documents are not sent to a model provider.
-
Reserved vCores and RAM
A model's memory is yours alone.
-
NVMe storage
Fast model loading.
-
Dedicated IPv4 and IPv6
A fixed address for your private API.
Why Host an AI Model on a VPS
Large hosted models are more capable, but every prompt leaves your control. A small model on your own server is a different trade-off.
-
Privacy by design
Sensitive text is processed on a server only you can reach.
-
A flat cost
No per-token charges for the models you run yourself.
-
Your own API
Applications call your server instead of a public service.
-
Freedom to choose
Pick and swap open models as better ones appear.
How a Self-Hosted Model Server Is Set Up
A model runtime, a private API and, if you like, a front end or a reverse proxy.
Request from a browser or app (HTTPS)
-
Runtime
Ollama
Runs open models on the CPU and exposes an API on localhost, including an OpenAI-compatible endpoint.
-
Models
Files on NVMe
Small, quantised open models that fit in RAM.
-
Access
SSH tunnel, WireGuard or reverse proxy
How your apps reach the API.
-
Front ends
Optional
A chat interface such as Open WebUI, or your own application.
Worth knowing
- The VPS plans have no GPU, so models run on the CPU: slower than on a GPU, and best with small models.
- A model must fit in free RAM with room for the context; the model's file size is a good first guide.
- The API has no login of its own: keep it private or put authentication in front.
- Open models have their own licences; check them before commercial use.
What Small Models Do Well
-
Summarising and rewriting short text
Notes, emails and short documents.
-
Classifying and tagging
Sorting tickets, comments or records.
-
Extracting fields from text
Pulling names, dates and amounts into structured data.
-
Embeddings for search
Vectors for your own retrieval tools.
-
Private drafting
A first draft you then edit yourself.
-
Prototyping
Build an AI feature before choosing a hosted model.
VPS Plans for AI Models
Choose by memory, not by CPU. VPS-2 (8 GB RAM) runs small models; VPS-4 (24 GB RAM) leaves room for larger ones. Expect slower replies than on a GPU.
VPS-1
Small websites, APIs, bots and development servers.
- 2 vCores
- 4 GB RAM
- 40 GB NVMe
- 500 Mbps
- Full root and SSH access
- Free initial stack installation
- Server hardening and firewall setup
- Daily backup of the previous 24 hours
- Unlimited traffic *
- Anti-DDoS protection
- Dedicated IPv4 and IPv6
- KVM virtualisation
VPS-2
Production apps, WordPress and online stores.
- 4 vCores
- 8 GB RAM
- 75 GB NVMe
- 1 Gbps
- Full root and SSH access
- Free initial stack installation
- Server hardening and firewall setup
- Daily backup of the previous 24 hours
- Unlimited traffic *
- Anti-DDoS protection
- Dedicated IPv4 and IPv6
- KVM virtualisation
VPS-3
Busy stores, SaaS products and several projects.
- 6 vCores
- 12 GB RAM
- 100 GB NVMe
- 2 Gbps
- Full root and SSH access
- Free initial stack installation
- Server hardening and firewall setup
- Daily backup of the previous 24 hours
- Unlimited traffic *
- Anti-DDoS protection
- Dedicated IPv4 and IPv6
- KVM virtualisation
VPS-4
High-traffic apps, large databases and agencies.
- 8 vCores
- 24 GB RAM
- 200 GB NVMe
- 3 Gbps
- Full root and SSH access
- Free initial stack installation
- Server hardening and firewall setup
- Daily backup of the previous 24 hours
- Unlimited traffic *
- Anti-DDoS protection
- Dedicated IPv4 and IPv6
- KVM virtualisation
* Unlimited traffic in Europe and North America. In Mumbai, Singapore and Sydney a monthly allowance applies (VPS-1 500 GB, VPS-2/VPS-3 1 TB, VPS-4 3 TB), then speed is limited to 10 Mbps until the next month.
Prices exclude 18% GST, which is added to invoices for customers in India.
Which Plan for Your Model
A quantised model needs about its file size in free RAM plus room for context. These are guides, not speed guarantees.
| Workload | Start with | Why |
|---|---|---|
| Tiny models (a few billion parameters or fewer) | VPS-1 2 vCores · 4 GB RAM | 4 GB RAM; classification and short tasks. |
| Small models (about 7 billion parameters) | VPS-2 4 vCores · 8 GB RAM | 8 GB RAM; one model at a time. |
| Small model plus an application on the same server | VPS-3 6 vCores · 12 GB RAM | 12 GB RAM and 6 vCores. |
| Larger models, or several small ones | VPS-4 8 vCores · 24 GB RAM | 24 GB RAM; big models answer slowly on CPU. |
A starting point, not a limit: every plan upgrades in place, keeping your data, IP address and configuration.
GPU Cloud, Hosted API or a CPU VPS?
Choose by what matters: speed and model quality, or privacy and a flat cost.
| Feature | Hosted model API | CPU VPS |
|---|---|---|
| Model quality | Large, state-of-the-art | Small open models |
| Speed | Fast | Slower; no GPU on our plans |
| Where prompts go | The provider's servers | Your server only |
| Pricing | Per token or per request | Flat server cost |
| Best for | Quality and scale | Privacy and low volume |
| See Plans |
Working With Your Model Server
Everyday commands after handover.
-
Download a small model
An example; choose your own.
-
Test it in the terminal
Type /bye to leave.
-
Use the OpenAI-compatible API
From the server itself.
-
Watch memory while it runs
Check that the model fits.
# 1. Download a small model
ollama pull llama3.2:3b
# 2. Test it in the terminal
ollama run llama3.2:3b
# 3. Use the OpenAI-compatible API
curl http://localhost:11434/v1/chat/completions -H "Content-Type: application/json" -d '{"model":"llama3.2:3b","messages":[{"role":"user","content":"Say hello."}]}'
# 4. Watch memory while it runs
free -h
ollama ps
Example commands; adjust names, paths and versions to your project. Deploying and maintaining your application is yours to do (or ask us for a quote).
Installed and Secured Before You Log In
The VPS is unmanaged: after handover the server is yours to run. Before that, our engineers do the first setup once, free, so you start from a hardened server with AI Models in place.
-
Choose a plan and location
Pick the resources you need and the datacentre closest to your users.
-
Tell us your AI Models setup
Versions, database and domain: at order or right after.
-
We install and secure it
Server hardened, AI Models installed and tested, HTTPS on your domain.
-
You take over with root
SSH access to a working server, ready for your code and data.
Your model server setup
- Ollama installed with its official installer and running as a service
- The API on localhost only, with the firewall open for SSH
- Docker installed on request, for front ends you add yourself
- No models downloaded: you choose them and read their licences
Yours to run after handover
- Ongoing server administration
- Application maintenance and updates
- Debugging your code
- Migrating existing sites or data
Not included in the free setup, but we can quote for it. No control panel by default: CloudPanel, HestiaCP or Virtualmin on request at no cost; cPanel and Plesk need their own licence.
Platform Specs and Locations
The same KVM platform on every plan; only the CPU, RAM, storage and port speed change.
| Virtualisation | KVM with reserved vCores and RAM |
|---|---|
| Storage | NVMe SSD on every plan |
| Network | Dedicated IPv4 and IPv6, 500 Mbps to 3 Gbps per plan |
| Operating systems | Ubuntu 26.04, 24.04 and 22.04 LTS · Debian 13, 12 and 11 · AlmaLinux · Rocky Linux · Fedora · FreeBSD |
| Access | Full root over SSH, key-based login |
| Backups | Automatic daily backup of the previous 24 hours; snapshots and longer retention on request |
| Protection | Always-on anti-DDoS mitigation |
| Availability | 99.9% uptime SLA |
Secured before we hand it over
- Non-root sudo user created for daily work
- SSH key login, password login for root disabled
- Firewall (UFW) open only for the ports you use
- Fail2ban blocking repeated login attempts
- Automatic security updates enabled
- Free Let's Encrypt SSL on your domain
- Timezone, swap and hostname configured
Asia-Pacific
- Mumbai
- Singapore
- Sydney
Europe
- London
- Limburg
- Roubaix
- Gravelines
- Strasbourg
- Milan
- Warsaw
North America
- Beauharnois
Availability per location changes with demand; we confirm your location before setup. In Mumbai, Singapore and Sydney each plan includes a monthly traffic allowance (VPS-1 500 GB, VPS-2 and VPS-3 1 TB, VPS-4 3 TB); beyond it, speed is limited to 10 Mbps until the next month.
Who Hosting an AI Model Suits
-
Teams with private text
Contracts, notes or records that must stay on your server.
-
Developers prototyping AI features
A private endpoint with no per-token bill.
-
Low-volume internal tools
Occasional requests rather than heavy traffic.
Security and Performance for Model Hosting
Security
- Keep the model API private; it has no authentication of its own.
- Use an SSH tunnel, WireGuard or an authenticated reverse proxy for remote access.
- Check each model's licence and acceptable use.
- Keep the runtime and the operating system updated.
Performance
- Choose the smallest model that does the task.
- Prefer quantised models; they use less RAM.
- Keep one model loaded to avoid reload time.
- Test with your real prompts before committing to a plan.
Questions Before You Deploy
Not covered here? Ask an engineer before you order.
Host a Small Model on Your Own Server
Choose a plan and we hand over a hardened server with the model runtime installed, ready for the models you pick.