brama detect¶
Detect local hardware capabilities: the GPU and its dedicated memory, RAM, CPU cores, and CUDA and Metal availability.
Invocation¶
brama detect [--json]The only option is --json.
Output and state¶
It prints GPU type and name, VRAM, RAM, CPU core count, and CUDA and Metal availability. With --json it prints the same reading as one document under resources. It recommends no model: which model a host serves is decided by the inference routes, not by its memory size. It performs no network request, reads no credential, writes no state, incurs no provider cost, and exits 0.
The GPU is read from system_profiler SPDisplaysDataType on Apple Silicon and from nvidia-smi elsewhere; each run is announced on stderr when it starts and when it ends. A run that cannot start or exits unsuccessfully names no GPU, and the log says why; VRAM is reported only when nvidia-smi states a number for it.
Refusals¶
There is no product-level refusal. Clap rejects unsupported operands or options before detection runs.