Ollama setup: local models, files and agent tasks

This Ollama tutorial explains how to run models locally through a desktop chat app and terminal. Jamie demonstrates the workflow on Windows and names Mac and Linux as download options. He says installation needs no account, but the agent features require a recent version; he gives no minimum version number.

Model choice depends on available memory. The speaker distinguishes graphics-card VRAM on Windows from shared RAM on Apple silicon Macs and gives rough sizing guidance rather than guaranteed hardware requirements. He recommends checking download sizes and the tools, vision and thinking tags before choosing a local LLM. Gemma 4 is his suggestion for smaller machines, with a size that fits their memory.

A coffee-roasting business supplies the examples: a product description, sales spreadsheet analysis, label feedback from an image and a pricing calculation. Terminal coverage includes downloading, running, listing and removing models, plus checking which models occupy memory.

The AI agent demonstration researches pricing and saves files after requesting command approval. Jamie warns that agent tasks take longer than chat and that smaller models can make more tool-use mistakes. He also explains the privacy boundary: web search requires sign-in and uses an online service, while cloud models send prompts to a server. He presents the cloud toggle as a way to keep Ollama offline. The final section briefly covers connecting models to other apps through Launch and a local API.