GPU servers, model stacks, local inference experiments, and agent workloads — everything running on-prem instead of in the cloud.
LOCAL AI / COMPUTE
Running Hermes Agent on an Intel Mac mini with GPU inference on Unraid. A restricted Linux account, Tailscale, an authenticated Nginx proxy, and nftables limit access to the Ollama backend.
LOCAL AI / COMPUTE
Local AI inference on Unraid 7.2.7 with a Ryzen 5 5600G, 32 GB DDR4, and an NVIDIA Tesla V100 PCIe 32 GB. The GPU runs at a 150 W power limit with a custom 3D-printed fan shroud.