Running Ollama Locally: Your Private AI Server in 5 Minutes

Why Local LLMs Matter

Cloud APIs (OpenAI, Claude, Gemini) are powerful but come with trade-offs:

• Privacy — Your data is sent to third-party servers
• Latency — Network round-trips add 500ms-2s per request
• Cost — High-volume inference becomes expensive

Ollama solves this by running open-source LLMs locally on your server. Zero external calls, instant inference, unlimited usage.

Installation (Linux/Mac)

One command:

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

Start the service:

ollama serve

That's it. Ollama is now listening on http://localhost:11434

Popular Models

• codellama:13b — Code generation, Python/JavaScript/Odoo Python. 4GB RAM, 200ms response
• mistral:7b — Fast general-purpose, good for summarization. 3.5GB RAM
• llama3:8b — Reasoning, explanation, creative tasks. 4GB RAM
• neural-chat:7b — Conversational AI, lightweight. 3GB RAM

Using Ollama from Python

import requests
import json

response = requests.post('http://localhost:11434/api/generate', json={
    'model': 'codellama:13b',
    'prompt': 'Write an Odoo model for customer tickets',
    'stream': False
})

result = response.json()
print(result['response'])

Real-World Odoo Use Case

Imagine a CRM that auto-classifies incoming emails:

• Email lands in Odoo mail.mail
• Server hook calls Ollama: "Classify this email: urgent? sales? support?"
• Response in 200ms, zero API cost, no data leaves your server
• Auto-assign to queue, set priority, create lead

Performance Benchmark

Latency comparison (request: "summarize this contract in 50 words"):

• Claude API: 1200ms (network + inference)
• Ollama mistral:7b: 180ms (local inference)
• Ollama on GPU: 80ms

For high-volume tasks (100 emails/day), Ollama pays for itself.