You can use models like qwen3.5 running on local hardware in ollama and redirect Claude to use the local ollama API endpoint instead of Anthropic’s servers.