Operate provider roles

Provider operators run separate computation roles:

pllm serve inference --config inference.json
pllm serve preparation --config preparation.json

The inference role consumes one-use prepared rows online. The trusted preparation role computes and pushes masked corrections before a response begins. Neither role is an application-facing Responses API or Chat Completions API endpoint.

These commands start processes only. They do not provision hosts, certificates, identities, monitoring, rollback protection, or non-colluding operators. Keep credentials in protected configuration or environment variables rather than command arguments.

Read the generated references for serve inference and serve preparation, then apply the deployment boundary and preparation pipeline.