A Latent Space episode titled "Coding From the Himalayas With an Offline AI Agent" sets out how to build a coding assistant that runs entirely on local hardware. The setup pairs OpenCode, an open-source coding agent, with Ollama, a tool for downloading and running large language models on a personal machine, and Qwen, an open-weight model family from Alibaba. The premise is practical. Developers working in remote locations, on unreliable connections, or under strict rules about where code and data can go still want AI help with programming tasks. Running the stack locally means prompts and source files do not need to be sent to a hosted service. Local setups generally trade that convenience for hardware demands, since capable models need substantial memory and processing power to run at usable speeds, and smaller open-weight models typically trail the largest hosted systems on complex work. For Australian businesses and professionals, the change is modest. Local models remove per-token API charges but shift the cost to hardware. Keeping code and prompts on-device can reduce exposure under the Privacy Act 1988, particularly the cross-border disclosure rules in APP 8, but it does not remove obligations for any personal information a team handles. Developers with sensitive client code or patchy connectivity may find it worth trying. For most others, it changes little.