A launcher and setup guide for running Anthropic's "Claude Code" with local models on Windows has been published. The configuration uses Ollama with open-source models instead of an Anthropic API key, and is shared on GitHub.
Local AI Coding · Open Source Tooling
Run Claude Code Locally on Windows — No API Key, No Cloud
A public repository connects Anthropic's Claude Code CLI to a local Ollama instance, letting the agentic coding tool run on open-source models like Qwen — offline, privacy-focused, and at zero API cost.
$0
API cost — runs fully offline on local models
1.7b
Default model (qwen3) tuned for low-spec machines
~min
Beginner-friendly setup in a few minutes
How the connection works
Claude Code CLI
@anthropic-ai/claude-code
→
Env vars
BASE_URL → localhost:11434
→
Local Ollama
Messages API compatible
→
Qwen model
reads · edits · runs
Launch via .\claude-ollama.cmd — needs Node.js, Ollama and the npm-distributed Claude Code installed. GPU crash? Disable it via CUDA_VISIBLE_DEVICES to fall back to CPU.
✓ Seen as practical
Privacy-focused, offline, zero API cost
Easy GPU→CPU troubleshooting on crash
Local codebase analysis & editing
Script automation, VS Code & WSL integration
△ The trade-offs
Small 1.7b–4b models give lower-quality code
Slower runs than official models
Unstable tool calls (tool_use / tool_result)
Some say llama.cpp direct is more flexible
Local execution buys cost and privacy — but the quality you get scales with the model you can run.
Accepting quality trade-offs based on model size remains the open challenge going forward.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…