Distributed AI Orchestrator

Multi-machine campaign orchestrator for coordinated AI workloads. Defines campaigns as collections of tasks with dependency graphs — parallel tasks fire simultaneously, dependent tasks execute in topological order with each wave waiting for its prerequisites to complete. Machine selection scores available nodes by capability match, available execution slots, memory headroom, and hardware class.

Task tracking with distributed state management. Each task carries its priority tier, capability requirements, and execution constraints. The orchestrator monitors running tasks per machine and routes new work to the node with the best fit — not just the least loaded, but the most capable for the specific workload type (LLM inference, image generation, multimodal analysis, code execution).

Campaign templates for common multi-step workflows: code analysis pipelines, research synthesis, content generation chains. Single-task dispatch available for one-off execution without campaign overhead. Results persist with full provenance — which machine executed each task, elapsed time, and output location.

Designed for workloads that span multiple machines and multiple stages — where a single node can't handle the full pipeline, and where failure at any stage needs to be recoverable without restarting the entire chain. The orchestrator handles the coordination logic so individual services only need to execute their specific capability.