Command line¶

The lightsmith command wraps the Python API. It is built for scripts and agents:

every command prints JSON to stdout;
every failure prints {"error": "<ExceptionType>", "message": "..."} to stderr and exits with code 1.

lightsmith doctor
lightsmith backends
lightsmith schema
lightsmith estimate SPEC [--method auto] [--device auto]
lightsmith solve    SPEC [--method auto] [--device auto] [-o result.json]
lightsmith compare  SPEC [--methods mie,dda] [--device auto] [-o comparison.json]

Command	Output
`doctor`	NumPy, PyTorch and CuPy versions, the CUDA GPUs, and the device `auto` would pick
`backends`	One entry per backend: `name`, `method`, `problem_kinds`, `license`, `gpu`, `differentiable`, `available`, `unavailable_reason`, `citation`
`schema`	The JSON Schema of every problem kind
`estimate`	Backend, device, problem size and memory, without solving
`solve`	The result model as JSON
`compare`	`reference`, `results` (one per method) and `relative_difference`

SPEC is a JSON file of any problem kind, or - to read from stdin.

Option	Values
`--device`	`auto`, `cpu`, `gpu`, `cuda:N` or `torch-cpu`
`--method`	`auto`, a method such as `dda`, or a backend name such as `lightsmith-mie` (see Solvers)
`--methods`	Comma-separated methods or backend names. Default: every backend that applies.

Examples¶

lightsmith estimate examples/drude_sphere.json --method dda --device gpu
lightsmith solve examples/drude_sphere.json -o result.json
cat spec.json | lightsmith solve - --method rcwa
lightsmith compare examples/drude_sphere.json --methods mie,dda,fdtd

The MCP server exposes the same operations to AI agents.