Console & fleet

Automatic model selection

In the device profile, model size can be set to Automatic (by hardware) instead of a fixed size. The device reports its hardware at check-in, and the server assigns the largest model size that hardware can run, capped by the booked plan.

~3 min read · Console & fleet

What it is

The owner picks Automatic (by hardware) as the model size in the device profile. From then on, every check-in re-reads the device's hardware and the server assigns the largest model size that hardware can carry, capped by the plan on the account.

Hardware to model mapping

HardwareModel
Dedicated GPU, ≥ 8 GB VRAM, with a runnable GPU engine, or Apple Silicon unified memory ≥ 16 GBUltra (GPU)
≥ 16 GB RAM with a capable CPUEnterprise
≥ 8 GB RAMPro
Below that, or a weak CPUNano

The GPU path today

Only Apple Silicon (Metal) qualifies for the GPU path today; support for dedicated GPUs (Vulkan) is coming. When hardware detection is uncertain, auto conservatively picks the best CPU tier and never picks GPU.

The plan caps the result

PlanHighest size auto can pick
ProPro
EnterpriseEnterprise or Ultra

Turning it on

  • In the device profile, set model size to Automatic (by hardware).
  • Optionally bind that profile to an install code: every device installed with the code then selects its model automatically, with no per-device step.
Check-in reports only coarse hardware facts: GPU type, video memory, system memory and CPU features. Never file or memory content.

Bundle model size with mode and second opinion: profiles for a fleet. The CPU baseline each size still needs: system requirements.

Was this helpful?

The right size, automatically.

Set it once in the device profile; every device then runs the largest model its hardware supports, capped by your plan.

Sign me up