Run Claude Code With Open Models Using Ollama

In short: Ollama lets you run Claude Code with open models on your own computer or in Ollama’s cloud. To be clear: the interface is Claude Code, but the model inside is not Claude. Local models cost nothing per token; quality is below Claude.

Quick setup (Windows)

1. Download and install Ollama from ollama.com.
2. If you don’t have Claude Code, install it in PowerShell: irm https://claude.ai/install.ps1 | iex
3. Pick a model that supports tool use (for example qwen3.5).
4. Start: ollama launch claude

Manual setup (PowerShell)

$env:ANTHROPIC_AUTH_TOKEN="ollama"
$env:ANTHROPIC_API_KEY=""
$env:ANTHROPIC_BASE_URL="http://localhost:11434"
claude --model qwen3.5

If your Claude Code settings define another address, it overrides these variables. Check the connection with /status.

Model and context

• The model must support tools.
• Set the context length to at least 64k, or jobs get cut off.
• Bigger models do better but need more RAM and GPU memory.

Cloud models

You can also connect to Ollama’s cloud without installing Ollama: the address is https://ollama.com and the key is your OLLAMA_API_KEY (example model: glm-5.3-flash). Check pricing and quotas on Ollama’s site first.

Good for, not good for

• Good: small fixes, simple scripts, explaining code, learning.
• Weak: big refactors, long agent runs, complex debugging.
• If the model keeps repeating a mistake or the job risks data loss, switch back to real Claude.

Free PDF guide

Setup steps, common errors and fixes in a 3-page guide (in Turkish).

Source: Ollama docs – Claude Code. Accurate as of November 14, 2026.