Your Own Private AI, Part 2: Secure Access from Anywhere with Tailscale Aperture (10io.com)

In the latest development of the "Your Own Private AI" series, Part 2 outlines steps to securely access the Qwen3.6-35B-A3B-FP8 Mixture of Experts model hosted on a GB10 NVIDIA DGX Spark-class machine from multiple devices without exposing it to the public internet. Leveraging Tailscale for a private, encrypted mesh network and its new AI gateway, Aperture, users can run their self-hosted AI models while maintaining privacy and security. This two-layered approach ensures that only authorized devices can communicate with the model, eliminating the risks associated with traditional port forwarding. This implementation is significant for the AI/ML community as it provides a blueprint for creating secure, accessible AI solutions that can rival commercial offerings from providers like OpenAI or Anthropic. The method includes technical specifics such as joining a tailnet, configuring the Aperture gateway, and authenticating requests based on Tailscale identities, which enhances security while simplifying access. By encapsulating private AI infrastructures within managed networks, users can explore applications without compromising their models' integrity or accessibility.