Provider roles

Provider supplies static component metadata and implementation artifacts. Operator supplies process placement, authenticated identity, authorization, storage, and operational policy. Preparation creates or installs method-specific material. Inference consumes assigned plan work. A deployment may combine processes, but cannot erase semantic role boundaries or establish non-collusion.

Metadata discovery through pllm components never loads providers or native libraries. Research discovery never executes recipes. pllm serve inference and pllm serve preparation start the two roles, but authenticated role-plan delivery and a complete target lifecycle contract remain blockers. See the serving guide.

Python SDK example

from pllm.components import get, get_component
from pllm.roles import Inference

role_class = get("pllm/inference")
method = get_component("pllm/masked-linear")
assert role_class is Inference
assert role_class.describe().role_eligibility == ("inference",)
assert method.role_eligibility == ("client", "preparation", "inference")

API: pllm.components.get and get_component

Role metadata does not establish operator identity, authentication, or non-collusion.