Skip to main content
Gateway job pipelines are the control-plane path between app requests and orchestrator execution. A gateway does not run model inference; it accepts requests, applies routing policy, selects compatible orchestrators, and returns results.

What a gateway pipeline does

  • Accepts AI requests from apps and clients
  • Matches request requirements to advertised orchestrator capabilities
  • Applies policy: price ceilings, latency expectations, retries, failover
  • Dispatches work to orchestrators and returns outputs

Gateway vs orchestrator responsibilities

LayerPrimary responsibility

Pipeline types

  • Real-time frame pipelines (video-to-video, overlays, style transfer)
  • Streaming audio pipelines (ASR and translation)
  • Capability-composed pipelines (for example, depth -> conditioning -> generation)

BYOC in gateway pipelines

Gateway operators can route to orchestrators that expose BYOC capabilities. BYOC extends what workloads a gateway can serve by allowing operators to run custom inference containers behind orchestrator capabilities.

See also

Last modified on March 2, 2026