Skip to content

Orchestrator

The Orchestrator page is the dashboard surface for the control-plane agent that routes and supervises work.

The orchestrator is not just another role. It is the part of the system that decides what should happen next, which is why it gets a dedicated surface instead of being buried under generic role settings.

  • orchestrator prompt baseline
  • model-routing posture for orchestrator work
  • pool posture and related execution defaults

Use this page when you need to tune how the orchestrator plans, routes, or supervises specialist work. The effects tend to show up across the whole product, not just in one role.

A small change here can alter how aggressively the system decomposes work, revisits blocked situations, or delegates tasks to specialists.

Because the orchestrator owns planning and recovery decisions, weak model choices show up here quickly. For the best current balance of quality and end-to-end speed, we recommend gpt-5.4 with at least low reasoning, and usually medium, rather than a low-capability or non-reasoning model.

The orchestrator sits between authored process and specialist execution:

This page answers “how should the control-plane brain behave?” The workflow surfaces answer “what did it decide on real work?”