In this video I show you how I’m using local models inside Claude Code completely free, using Ollama. I demo it by building a real-time weather app that hits a live API, pulls accurate data for any city, and wraps it in nice UI animations—while the actual coding agent runs locally on my machine. No subscriptions, no API costs, and you can run securely offline. The best part is you still get the full Claude Code experience: slash commands, MCP servers, sub-agents, and switching between plan/edit modes—everything just works on consumer-grade hardware. I walk you through installing Ollama, pulling models, and then launching Claude Code the “special” way so it doesn’t ask for an Anthropic login or API key. I also break down which models to use and why: GPT-OSS 20B is the smallest, Qwen3-Coder is my current favorite (better results for me), and GLM 4.7 Flash is a solid option—especially if you’ve got the VRAM for the bigger quant. I’ll also cover troubleshooting (like 404s or version mismatches) and how context window sizes differ, because that directly affects response quality once your sessions get large.