utsav@homelab:~$
cd ~/homelab/ai-server
LOCAL AI SERVER
> U6 · NPU, A CHATGPT STYLE INTERFACE THAT NEVER LEAVES THE HOUSE
utsav@homelab:~/homelab/ai-server$ cat README.md
A Mac mini M4 set up as a dedicated, always on local AI server alongside the Pi based homelab.
Ollama runs local language models with GPU acceleration through Apple's Metal backend, Open WebUI
gives it a familiar browser based chat interface, and a Google Gemini connection sits in the same
dropdown for anything that benefits from a larger cloud model. Everything is reachable from any
device on the home network or Tailscale, with a strict rule about what goes to which model.
utsav@homelab:~/homelab/ai-server$ ./features --list
▸ Four local models, general chat, coding, and two multimodal sizes
▸ Google Gemini connected alongside the local models in one dropdown
▸ External NVMe SSD for model storage, kept off the internal drive
▸ Auto starting on boot through launch agents and a launch daemon
▸ Caddy reverse proxy, clean local domains instead of IP and port numbers
▸ A simple privacy rule, local models for anything sensitive, cloud for the rest
utsav@homelab:~/homelab/ai-server$ ollama list --format stack
HARDWAREMac mini M4, 16GB unified memory, external NVMe enclosure
INFERENCEOllama, GPU accelerated through Apple Metal
INTERFACEOpen WebUI, browser based, run as a network server
MODELSqwen2.5, qwen2.5 coder, gemma4 (two sizes, multimodal)
CLOUDGoogle Gemini, free tier, connected as an extra provider
utsav@homelab:~/homelab/ai-server$ git log --reverse --milestones
2026-06-23Set up from scratch in one session. External SSD mounted, Ollama installed, four production models pulled after a quick test model was removed
2026-06-23Open WebUI running as a proper network server rather than the desktop only build, reachable from every device on the Tailnet
2026-06-23Caddy reverse proxy and local DNS records added, so the server is reached by name instead of an IP and a port
2026-07-05Bulk document ingestion tested at scale, a resource limit bug found and fixed along the way
ONGOINGDaily driver for coding help, quick questions and document search across both local and cloud models
utsav@homelab:~/homelab/ai-server$