TURBINEDOCUMENTATION

CONCEPTS

Understanding the core abstractions that power Turbine.

§ RIG

A Rig is a workspace containing one or more AI agents with shared configuration. Rigs provide isolation and resource allocation boundaries.

rig: protocol ├─ agent-0 (claude) ├─ agent-1 (claude) └─ agent-2 (codex)

Each rig can have its own rate limits, cost budgets, and provider preferences.

§ CONVOY

A Convoy is a sequence of related tasks executed across multiple agents. Convoys track progress from initiation to completion.

convoy: auth-system status: in_progress progress: 67% (8/12 beads) agents: [agent-0, agent-1]

Convoys enable high-level task tracking and cost attribution.

§ BEAD

A Bead is an atomic unit of work. Beads are the smallest trackable unit in Turbine, representing individual tasks or requests.

bead: bd-a2f3 type: implementation status: complete agent: agent-0 tokens: 4,521 cost: $0.14

Beads flow through the system from creation to completion, tracking all associated costs.

§ PROVIDER

A Provider is an AI model backend. Turbine supports multiple providers and handles routing, failover, and load balancing.

  • Claude (Anthropic)
  • Codex (OpenAI)
  • Gemini (Google)
  • Custom (self-hosted or other APIs)

§ RATE LIMITING

Turbine maintains a sliding window of API usage across all providers. When a provider approaches its limit, requests are automatically routed to available alternatives.

provider claude: 73% capacity (730/1000 rpm) provider codex: 45% capacity (225/500 rpm) provider gemini: 12% capacity (240/2000 rpm)

§ PRIORITY ROUTING

Tasks can be assigned priority levels that influence provider selection:

  • Critical — Always uses primary provider
  • High — Prefers primary, falls back if needed
  • Normal — Uses optimal available provider
  • Low — Routes to cost-effective alternatives