Edge AI

Intel NUC + Arc GPU

paid

Budget Edge AI for clients who won't buy NVIDIA.

— Intel's Arc GPU via OpenVINO is viable; performance varies by model family.

NVIDIA Jetson Orin Nano

paid

8GB, ~€250, runs Llama-3.2-8B class workloads.

— Our default Edge AI recommendation for cost-conscious deployments.

Mac Mini M4

paid

~€700, surprisingly competitive when macOS is acceptable.

— MLX + unified memory means you get throughput well above the price suggests.

llama.cpp

OSS

The C++ runtime that Ollama wraps.

— Go direct when you need quantization control Ollama hides.

MLX

OSS

Apple's on-device inference framework.

— The fastest path to good throughput on Apple Silicon; often beats llama.cpp for the same quantization.

OpenVINO

OSS

Intel's answer for their hardware.

— Worth a look if the client's stack is Intel-heavy; otherwise MLX + llama.cpp covers more ground.

systemd + LaunchAgents

OSS

OS-native service managers on Linux / macOS.

— How we ship Edge deployments. See VORLUX's `com.vorlux.*` agents in the platform.

Tailscale

freemium

Mesh VPN for remote management.

— What we use to reach client Edge nodes without opening ports. Free tier works.