ollama

Ollama: Déploiement IA Local

Exécutez n'importe quel modèle d'IA localement avec Ollama. Guide complet pour configurer Ollama avec AI Gateway comme solution de secours.

1 octobre 2026ollamalocalauto-hebergementhors ligne

Get API Key View Pricing

Ollama: Local AI Deployment

Ollama lets you run AI models locally on your machine — zero API costs, complete data privacy, offline operation. Pair Ollama with AI Gateway as a fallback layer: local models as primary, cloud as automatic fallback. Best of both worlds.

Models You Can Run Locally

Family	Popular Models	Active Params	Hardware
DeepSeek	V4-Pro, V4-Flash, R1	13B–49B	RTX 4090 or better
Llama	4, 3.3, 3.2	7B–70B	8GB–48GB VRAM
Qwen	3.8, 3.5, 3	0.6B–235B	4GB+ VRAM
Mistral	Small 3, Nemo, 7B	7B–22B	8GB+ VRAM
Gemma	4, 3, 2	2B–27B	6GB+ VRAM

Quick Start

Install

bash

# macOS / Linux
curl -fsSL https://ollama.com/install.sh | sh

# Windows: Download from https://ollama.com/download

ollama --version

Pull & Run

bash

# Pull models
ollama pull deepseek-v4-flash
ollama pull llama-3.3-70b
ollama pull qwen-3.5-flash

# Run interactively
ollama run deepseek-v4-flash

# Run with custom context
ollama run deepseek-v4-pro --num_ctx 131072

Python Integration

python

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:11434/v1",
    api_key="ollama"  # Required but unused
)

response = client.chat.completions.create(
    model="deepseek-v4-flash",
    messages=[{"role": "user", "content": "Explain Docker in simple terms."}]
)

Key Capabilities

Zero API Costs

After hardware purchase, running models costs nothing per query

No per-token charges
No rate limits
Unlimited usage
One-time hardware investment

Complete Data Privacy

All data stays on your machine — HIPAA, GDPR compliant by design

No data leaves your infrastructure
Full audit trail locally
Enterprise security requirements met
Healthcare and legal use cases

Offline + AI Gateway Fallback

Seamless hybrid: local models primary, cloud as automatic fallback

AI Gateway detects Ollama availability
Automatic failover when offline
No user-facing errors
Privacy + reliability combined

Hybrid Setup

python

import socket

def is_ollama_running():
    try:
        sock = socket.socket()
        sock.settimeout(1)
        sock.connect(("localhost", 11434))
        sock.close()
        return True
    except:
        return False

def get_client():
    if is_ollama_running():
        return OpenAI(base_url="http://localhost:11434/v1", api_key="ollama")
    else:
        return OpenAI(api_key="your-ai-gateway-key", base_url="https://api.ecaix.com/v1")

# Seamlessly switches between local and cloud
client = get_client()
response = client.chat.completions.create(
    model="deepseek-v4-flash",
    messages=[{"role": "user", "content": "Hello."}]
)

Hardware Guide

Model Size	Min RAM	Recommended	GPU
7B	8 GB	16 GB	Optional
13B	16 GB	24 GB	Recommended
33B	32 GB	48 GB	Strong
70B	64 GB	128 GB	Required
405B	256 GB	512 GB	Multi-GPU

DeepSeek V4 — Best Ollama candidate (open weights, low compute)
Meta Llama — Largest ecosystem via Ollama
Qwen — Best multilingual via Ollama
Mistral — Fast small models for Ollama
View all models →

Ready to build with Ollama?

One API Key, every model. Transparent pay-as-you-go pricing with credits that never expire.

Create free account

Family

Popular Models

Active Params

Hardware

DeepSeek

V4-Pro, V4-Flash, R1

13B–49B

RTX 4090 or better

Llama

4, 3.3, 3.2

7B–70B

8GB–48GB VRAM

Qwen

3.8, 3.5, 3

0.6B–235B

4GB+ VRAM

Mistral

Small 3, Nemo, 7B

7B–22B

8GB+ VRAM

Gemma

4, 3, 2

2B–27B

6GB+ VRAM

Quick Start

bash

# macOS / Linux
curl -fsSL https://ollama.com/install.sh | sh

# Windows: Download from https://ollama.com/download

ollama --version

bash

# Pull models
ollama pull deepseek-v4-flash
ollama pull llama-3.3-70b
ollama pull qwen-3.5-flash

# Run interactively
ollama run deepseek-v4-flash

# Run with custom context
ollama run deepseek-v4-pro --num_ctx 131072

python

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:11434/v1",
    api_key="ollama"  # Required but unused
)

response = client.chat.completions.create(
    model="deepseek-v4-flash",
    messages=[{"role": "user", "content": "Explain Docker in simple terms."}]
)

Key Capabilities

After hardware purchase, running models costs nothing per query

No per-token charges

No rate limits

Unlimited usage

One-time hardware investment

All data stays on your machine — HIPAA, GDPR compliant by design

No data leaves your infrastructure

Full audit trail locally

Enterprise security requirements met

Healthcare and legal use cases

Seamless hybrid: local models primary, cloud as automatic fallback

AI Gateway detects Ollama availability

Automatic failover when offline

No user-facing errors

Privacy + reliability combined

Hybrid Setup

python

import socket

def is_ollama_running():
    try:
        sock = socket.socket()
        sock.settimeout(1)
        sock.connect(("localhost", 11434))
        sock.close()
        return True
    except:
        return False

def get_client():
    if is_ollama_running():
        return OpenAI(base_url="http://localhost:11434/v1", api_key="ollama")
    else:
        return OpenAI(api_key="your-ai-gateway-key", base_url="https://api.ecaix.com/v1")

# Seamlessly switches between local and cloud
client = get_client()
response = client.chat.completions.create(
    model="deepseek-v4-flash",
    messages=[{"role": "user", "content": "Hello."}]
)

Model Size

Min RAM

Recommended

GPU

8 GB

16 GB

Optional

13B

16 GB

24 GB

Recommended

33B

32 GB

48 GB

Strong

70B

64 GB

128 GB

Required

405B

256 GB

512 GB

Multi-GPU