Ollama's `ollama launch claude` sets up Claude Code with an open model (like Qwen) running on your own machine: no Claude subscription, no API key, no config files. For coding, raise Ollama's context length to 64K; no big GPU? `:cloud` models run on Ollama's servers (small free monthly allowance on starter models, then pay per token). Honest note: open models are weaker than Claude on hard tasks, and Anthropic doesn't support Claude Code on non-Claude models.