Self-host a local AI stack and access it from anywhere

Running AI locally is becoming practical and private thanks to new models like Llama 4, Gemma 3, and DeepSeek, allowing users to control their data without monthly fees. The process involves setting up a local AI lab using tools such as Proxmox for virtualization, NixOS for system repeatability, Docker for packaging, and Tailscale for secure remote access. A compatible hardware setup, including an NVIDIA A4000 GPU and appropriate CPU support, is necessary for PCIe passthrough. The setup creates an "AI Basecamp" with a NixOS VM equipped with Docker, NVIDIA drivers, and Tailscale, enabling the use of Ollama and Open WebUI for a local ChatGPT-like experience. This system, accessible securely from anywhere via Tailscale Serve, can be automated with Ansible for easy reuse and the safe integration of secrets. It allows for offline LLM use, facilitating projects like Home Assistant integrations and code generation while also enabling cost-sharing of GPU resources for private AI access among friends. Comprehensive guidance and resources are available, including a detailed walkthrough video and GitHub repository.