the setup · comment "CODE" drop

Claude Code on a free local model. The exact setup.

Three steps. No API bill, no internet needed, and your code never leaves your machine.

The honest part first: a local model is not Claude. It will not match the paid models, and you will feel it the moment a task needs several files at once. Use this for the routine work, keep the subscription for the hard problems.

The setup

STEP 1

Install Ollama

Free app that runs AI models on your own computer. Windows, Mac, Linux.

ollama.com ↗

STEP 2

Grab a coding model

Qwen 2.5 Coder is the sweet spot for consumer laptops. It fits in about 4 GB.

ollama pull qwen2.5-coder

STEP 3

Point Claude Code at your own machine

Claude Code reads one setting that decides which server it talks to. Change it in your settings.json:

"env": {
  "ANTHROPIC_BASE_URL": "http://localhost:11434",
  "ANTHROPIC_MODEL": "qwen2.5-coder",
  "ANTHROPIC_AUTH_TOKEN": "ollama"
}

Ollama speaks the Anthropic API format natively now, so no proxy is needed. Restart Claude Code and it runs on your laptop.

What to use it for

Good, free: renaming things across a file, writing a single function, explaining code you didn't write, small fixes.
Keep paying for: big refactors, multi-file features, anything where it has to hold five files in its head at once. Quality isn't cheap.

The setup

Install Ollama

Grab a coding model

Point Claude Code at your own machine

What to use it for

Questions about this? Ask in the Discord.