Español
Linux Essentials
2 vCPU · 4 GB RAM
4,99 €/month
i.e. 3,99 €/month billed yearly
- vCPU2 Xeon cores
- RAM4 GB
- NVMe20 GB
- IPv41 dedicated
- Anti-DDoSGCore included
Ollama VPS: a private AI model on your own server
Run open models such as Llama, Mistral, Qwen or Gemma with Ollama on a Linux VPS in Paris. Your data stays with you, with no per-request billing. Without a GPU, aim for 3 to 8 billion parameter models for background tasks.
- Confidential data
- Available 24/7
- Full root access
- Fixed cost
ExcellentTrustpilot
GCore Anti-DDoS
GCore Anti-DDoSUp to 200 Tbps, included
Intel Xeon GoldServer-grade hardware and NVMe drives
Digital Realty ParisPAR6 datacenter, main site
Rated 4.4/582 reviews on Trustpilot
Choose the configuration for your Ollama VPS
Linux or Windows, five configurations each. We recommend Linux Business for this use. Prices are kept up to date from our catalogue.
These tools run very well on Linux, which is lighter and cheaper: that is our recommendation. Windows plans are there if you need a remote desktop.
No commitment
GCore Anti-DDoS included Automatic delivery
Prices excl. VAT, billed monthly
Hosting a VPN is not allowed on our VPS plans, except with the paid WireGuard module or our VPN offer.
Pre-installed when you orderSelect the application when you order: it installs itself on your VPS.
All plans are also on the pages Linux VPS and Windows VPS.
What you get
01
Confidential data
Text sent to the model never leaves your server, hosted in France.
02
Available 24/7
Your Ollama API answers your workflows and scripts, even with your PC off.
03
Full root access
Docker, Open WebUI, n8n or your own code: install whatever you need.
04
Fixed cost
One monthly price, however many requests you process.
Online in three steps
1
Choose the memory
8 GB for a 3 billion parameter model, 16 GB for 7 to 8 billion, 32 GB for 14 billion.
2
Install Ollama
One Docker command, or one-click installation from our marketplace, with Open WebUI if you want a chat interface.
3
Connect your tools
n8n, your Python scripts or a web interface call the local API, which is never exposed directly to the Internet.
What a CPU-only VPS does well, and what it does not
On a CPU, a 3 billion parameter model answers at a comfortable reading speed, around 15 to 30 tokens per second according to published measurements. A 7 to 8 billion model produces about 5 to 15 tokens per second: ideal for sorting emails, extracting data, summarising documents or computing embeddings, less so for real-time chat with several users.
A local model makes sense for three reasons: data confidentiality, a large volume of small repetitive tasks that would inflate an API bill, or independence from a provider. For complex reasoning, large models reached through an API remain far ahead, so many setups combine both: local for sorting and sensitive data, API for the rest.
Ollama has no authentication: keep its port bound to 127.0.0.1 or the internal Docker network, and go through Open WebUI, an SSH tunnel or a protected proxy. An exposed instance can be used by anyone.
Ideal for
- 1Sorting and tagging emails or tickets
- 2Extracting JSON data from documents
- 3Overnight summaries and document search (embeddings)
- 4Processing sensitive data without an external provider
Ollama on a VPS- Data stays on your server
- Fixed cost, no per-request billing
- Complex reasoning and advanced code
- Fast answers for many users
Large model API
- Data stays on your server
- Fixed cost, no per-request billing
- Complex reasoning and advanced code
- Fast answers for many users
Servers close to your users
Notre point de présence principal est en France, chez Digital Realty à Paris (PAR6). Nos VPS Linux sont aussi disponibles en Allemagne, au Royaume-Uni, aux États-Unis et au Canada.
FranceDigital Realty Paris (PAR6) · main PoP
Available
GermanyCentral Europe
Available
United KingdomLondon area
Available
United StatesNorth America
Available
CanadaNorth America
Available
+
New points of presence are comingListed here as soon as they open
Soon4,4/5
Our customers say it better than we do
- Automatic delivery
- No hidden fees
- French-speaking support
- No commitment
Frequently asked questions
Everything you need to know before ordering your Ollama VPS.
Another question?
Our team answers by ticket or chat.
Contact supportCan Ollama run without a GPU?
Yes, on the CPU, with quantised models. Models of 3 to 8 billion parameters are the realistic range, better suited to background tasks than real-time chat.
How much RAM does Ollama need?
About 8 GB of RAM for a 3 billion parameter model, 16 GB for 7 to 8 billion and 32 GB for 14 billion, plus 2 to 4 GB if n8n or a database run on the same server.
Which model should I start with?
A 3 billion parameter model such as Llama 3.2 3B to test the speed, then a 7 to 8 billion model if quality is not enough and you have 16 GB.
Can I get an interface like ChatGPT?
Yes, with Open WebUI, which can be installed when you order. Protect it with a password and HTTPS, and keep Ollama itself unreachable from the Internet.
Is my data sent to a third party?
No. The model runs on your VPS in Paris: the text you send it stays on your server.
Does Ollama work with n8n?
Yes. In n8n, an Ollama credential points to the container's internal address, and the Ollama Chat Model node serves as the model for your agents or chains.
Guides to get started
Our step-by-step tutorials to set up and secure your Ollama VPS.
Ready to launch your Ollama VPS?
Order in a few clicks, automatic delivery, no commitment.