How orchestration works
The Queen is the heart of SwarmADE. She runs server-side as a spec-driven orchestrator: you write what you want in plain language, submit it from your Hive, and the Queen turns it into a dependency-ordered plan, drives it to completion across your local Workers, and audits every result before it lands.
Where the Queen lives
The Queen runs in the cloud, not on your machine. That keeps planning and orchestration consistent regardless of your laptop, while the actual code-writing happens locally in your Workers — the CLI agents you already trust. Your Swarm is the set of Workers active on a given task.
The orchestration loop
You submit a spec; the Queen plans, dispatches, gathers, reviews, and advances until the plan is done.
spec ──▶ Queen (plan: dependency-ordered task graph)
│
▼
dispatch task ──▶ local Worker (your Swarm)
▲ │
│ ▼
advance plan ◀── Drone review (server-side audit)
│
▼
all tasks doneAt each turn the Queen picks the next ready tasks, sends them to available Workers, collects their output, hands completed work to the Drones for review, and only then advances the plan. Orchestration is hierarchical: the Queen can decompose a task further when it turns out to be bigger than it first looked.
Every Queen planning pass and every Drone review consumes Honey. Worker CLI calls run on your own subscription and cost no Honey. You can watch the running cost live while a plan executes.
Read next
- Planning a spec — how a spec becomes a dependency-ordered task graph.
- Execution loop — how the Queen drives tasks across the Swarm and gates them on review.
- Replanning + retries — what happens when work fails or reality diverges from the plan.
The Queen never edits your files directly — she orchestrates the Workers that do.