// Personal Project · Infrastructure
Home Lab & Remote Infrastructure
Set up local LLMs and coding agents on a home GPU server, reachable anywhere over an encrypted Tailscale mesh.
My home lab is a desktop with an RTX 5090 that runs local LLMs and coding agents, plus a Raspberry Pi that stays on all the time. Tailscale links them to my laptop and phone, so I can use all of it from anywhere.
Network Architecture
Every device is on a private Tailscale network, so they reach each other directly wherever they are, without opening any ports to the internet. The Pi also works as an exit node, so on public Wi-Fi my traffic can go out through home.
GPU desktop
Ubuntu with an RTX 5090. It runs local models with llama.cpp and Hermes, the coding agents, and containerised services in Docker.
Raspberry Pi
Low power and always on. It runs some of the containerised services and is the network's exit node.
Local Inference
Open-weight models run on the 5090 with llama.cpp, tuned for long inference runs. Prompts and data stay on my own hardware, and because the devices talk to each other directly over Tailscale, using a model from my laptop or phone feels local.
Collaborative Research Workspace
Friends working on research projects can use the 5090 too. Each person gets SSH access straight into a shared project folder, and nothing else.
Granular Access Control
Linux access control lists let collaborators into the shared folder while keeping every home directory private.
Automated Onboarding
A bash script sets up each person's SSH keys and environment.