What is Revue?

Revue makes Amp swarms better at asking humans for judgment.

Install Revue, connect one workspace-scoped API key, and run revue. Revue launches one Amp manager session that can supervise local children, remote Orbs, and subagents. Amp keeps doing normal coding work, but Revue adds a decision loop around the swarm: search prior decisions, search shared context, price whether a human question is worth asking now, route the answer back into the right thread, and remember the answer for the next task.

What users get immediately

Need	Revue behavior
Fewer repeated questions	Prior human decisions are searched before a new bid is allowed
Better interruptions	Human asks include risk, options, recommendation, evidence, and affected files
Thread steering without copy/paste	Human decisions route back into the owning Amp thread
Shared context across work	Agents can publish and query radio notes without spending attention
Proof the answer landed	Routes progress through delivered, consumed, applied, resolved, or failed
Swarm supervision	The manager thread can launch children/Orbs, list lineage, and answer decisions with delegation proof

Product scope

The active product is the Amp manager session backed by the hosted Revue API. Bare revue is the launcher; the local TUI remains a fallback UI.

Linear and Centaur code remains in the repository as adapters, but the engineer-first product language is Amp manager sessions, child sessions, Orbs, subagents, decisions, routes, and proof.

Mental model

Amp manager thread
  → launches children, Orbs, and subagents
  → agents search memory and shared context
  → agents bid only when judgment is worth interrupting for
  → human answers once in manager chat
  → Revue routes the decision back into the right thread
  → answer becomes reusable memory

Start with the Quickstart, then read the Amp plugin API.