Skip to main content

Runtime Core

Runtime Core is the execution layer for SLMCortex.

It loads a runtime bundle, validates it, picks the right route for a request, and either performs inference or returns the dry-run route that would have been used.

What It Does​

  • validates runtime bundles before use
  • normalizes inference requests
  • selects routes from the composed bundle
  • loads the base model plus selected adapters for real inference
  • exposes the same logic through CLI inference and the compatibility server

What Goes In​

  • one runtime bundle
  • a prompt or OpenAI-style request payload
  • optional overrides such as task_type, semantic_family, and slm_override

What Comes Out​

  • validation results from validate-runtime
  • route details from infer --dry-run
  • generated completions for real inference or server requests

Why It Matters​

Runtime Core is the layer that makes the packaged system useful instead of just well-organized. It is where routing, validation, and execution finally meet.

Where It Sits In The Flow​

validate-runtime, infer, and serve live here.

The agent layer sits on top: Agent Runtime.