Skip to main content

Future LLM Runtime (WIP)

This page describes a future runtime direction, not a current implementation.

Why this exists​

The operator model is designed so different engines and runtimes can sit behind the same contract. An LLM-backed runtime is one possible future direction.

How it would work​

A future LLM runtime would likely provide:

  • model access
  • tool access
  • execution memory for the run
  • structured result production

The engine would still be the executor. The runtime would still be resources and state.

Example​

A future run might look like:

await operator.execute({
  goal,
  capability,
  context,
  runtime: {
    model,
    toolset,
  },
  verify,
});

Notes​

WIP

There is no full LLM runtime implementation in this repo today.

Common confusion

An LLM runtime would still not change what a capability is. Capability is task-domain meaning. Runtime is execution resources.