How much RAM does Ollama need?

A 7B parameter model in Q4 quantization requires roughly 5–6 GB of RAM. Plan for 8 GB minimum to run a 7B/8B model comfortably. For 13B models, 16 GB; for 70B models (slow on CPU), 64 GB or a GPU is recommended.

Is Ollama compatible with the OpenAI SDK?

Yes. Ollama exposes an OpenAI-compatible endpoint at `/v1/chat/completions`, `/v1/completions`, and `/v1/embeddings`. Change `base_url` to your Ollama host and set `api_key` to any non-empty string — existing code requires no other changes.

Does Ollama work without a GPU?

Yes. CPU-only inference works out of the box. Smaller quantized models (Q4_K_M) run at usable speeds on modern CPUs for dev or low-volume use. For production latency on 7B+ models, a GPU accelerates inference 10–30×.

How do I add more models?

Run `docker exec -it ollama ollama pull ` from your VPS. Browse available models at ollama.com/library. Popular choices: `llama3.1:8b`, `mistral:7b`, `qwen2.5:7b`, `deepseek-r1:7b`, `gemma2:9b`.

Can I pair Ollama with Open WebUI?

Yes — they are designed to work together. Open WebUI auto-detects a local Ollama instance and lists all pulled models in its model selector. Both can run on the same VPS; Open WebUI adds a full chat UI, user management, and RAG on top of Ollama's inference engine.

Créer, héberger et exploiter des solutions IA.

Ollama

Run open-source LLMs on your VPS — OpenAI-compatible API, zero per-token cost.

8 Go minimum (16 Go recommandé) RAM 2 vCPU Port 11434 Disponible

Code source

Stack technique

DockerGollama.cpp

RAM minimum8 Go minimum (16 Go recommandé)

CPU minimum2 vCPU

Port par défaut11434

OS compatiblesubuntu-24.04

Ollama is an open-source LLM runtime that lets you pull and serve large language models (Llama 3, Mistral, Qwen, DeepSeek, Gemma) on your own infrastructure with a single command. It exposes an OpenAI-compatible REST API on port 11434, so any tool or library that works with OpenAI also works with your self-hosted instance — without changing a single line of code.

Deployed on a ServOrbit VPS, Ollama becomes your private AI inference backend: unlimited requests at a fixed monthly cost, with all prompts and responses staying on your server. It pairs directly with Open WebUI for a chat interface, Flowise or n8n for automation, or any custom application that calls `/v1/chat/completions`.

Fonctionnalités clés

OpenAI-compatible API — swap the base URL and your existing code works instantly

One-command model management: `ollama pull llama3.1:8b`, `ollama list`, `ollama rm`

Supports Llama 3, Mistral, Qwen, DeepSeek, Gemma, Phi and 100+ other models

CPU-only mode works out of the box; optional GPU passthrough for faster inference

Multi-model server: load several models concurrently, each on its own port or context

Persistent model storage in a Docker volume — models survive container restarts

Quand utiliser cette solution ?

1

Private AI backend

Replace OpenAI API calls with your own Ollama endpoint. Your prompts never leave your VPS — ideal for processing contracts, internal documents, or any sensitive data.

2

Local dev & testing

Run a 7B or 13B model on a dedicated VPS to prototype AI features before wiring up a production LLM. Fixed cost, no surprise bills, reset with `ollama rm`.

3

Shared team AI stack

Pair Ollama with Open WebUI to give your whole team a shared ChatGPT-like interface powered by open-source models — on infrastructure you control.

Déployer Ollama sur votre VPS

Guide optimisé pour les VPS Cloud ServOrbit.

01

Create the VPS

Order a ServOrbit VPS with at least 8 GB RAM (for a 7B/8B Q4 model in CPU mode) or 16 GB for larger models. Ubuntu 22.04 is recommended. For production-grade latency, a VPS with a GPU changes the experience significantly.

02

Deploy Ollama

Connect via SSH and run: docker run -d -v ollama:/root/.ollama -p 127.0.0.1:11434:11434 --name ollama --restart always ollama/ollama. Binding to 127.0.0.1 keeps the API off the public internet — it will sit behind your reverse proxy.

03

Pull your first model

Download a model: docker exec -it ollama ollama pull llama3.1:8b. Check it's running: docker exec -it ollama ollama list. Test locally: curl http://127.0.0.1:11434/api/generate -d '{"model":"llama3.1:8b","prompt":"Hello"}'.

04

Add a reverse proxy with auth

Ollama has no built-in authentication. Use Caddy or Nginx to terminate TLS and add a Bearer-token layer before exposing the API externally. Example Caddyfile: reverse_proxy localhost:11434 { header_up Authorization "Bearer {env.OLLAMA_TOKEN}" }.

05

Connect your tools

In Open WebUI → Settings → Connections, set the Ollama URL to http://localhost:11434. In any OpenAI SDK, set base_url='http://:11434/v1' and api_key='ollama'. Your tools now use your models.

Questions fréquentes

Ollama is an open-source runtime that downloads and serves large language models (LLMs) via a local REST API. Think of it as a local inference server: you pull a model once, and any application can call it just like it would call the OpenAI API.

Solutions complémentaires

Intelligence Artificielle

Activez Ollama sur votre infrastructure.

VPS Cloud dédié — IPv4 incluse, datacenter européen, support inclus. Vos données ne quittent jamais votre serveur.

Configuration recommandée : 8 Go minimum (16 Go recommandé) RAM · 2 vCPU