skypilot-multi-cloud-orchestration

Orchestrate multi-cloud ML workloads with cost-aware cluster selection and recovery.

Updated Feb 15, 2026
One-click install
npx skills add https://github.com/ricable/mcai --skill skypilot-multi-cloud-orchestration-ricable
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: skypilot-multi-cloud-orchestration
Source: https://github.com/ricable/mcai/tree/main/.agents/skills/skypilot-multi-cloud-orchestration
Command: npx skills add https://github.com/ricable/mcai --skill skypilot-multi-cloud-orchestration-ricable

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

This Skill removes the complexity of launching and managing ML workloads across multiple clouds, so you can focus on the job instead of cloud-specific setup, cost tradeoffs, and recovery handling.

Core Features & Use Cases

  • Multi-cloud launch planning: Choose the best cloud, region, and accelerator combination for a training or batch workload.
  • Cost-aware execution: Use spot instances, fallback clouds, and auto-selection to reduce GPU spend without sacrificing reliability.
  • Production ML operations: Run distributed training, checkpointed managed jobs, and scalable model serving with a unified SkyPilot workflow.
  • Use Case: A team needs to train an A100-based model, keep costs low, and survive preemptions; this Skill helps configure the task, storage, and recovery strategy end to end.

Quick Start

Ask the assistant to create a SkyPilot task for your ML workload with the cloud, GPU, and command you want, and it will generate a launch-ready plan.

Frequently Asked Questions about skypilot-multi-cloud-orchestration

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run distributed training across AWS, GCP, and Azure without manual setup?

Multi-cloud distributed training is orchestrated using a unified SkyPilot task YAML. You define accelerator and region constraints, and the system handles cloud-specific setup and execution across AWS, GCP, and Azure automatically.

What is the best way to reduce GPU spend for ML workloads using spot instances?

To reduce GPU spend with spot instances, use cost-aware execution with automatic fallback clouds and recovery. This approach minimizes costs while maintaining reliability through checkpointing and managed jobs during preemptions.

Can I use SkyPilot for scalable model serving on Kubernetes?

Yes, SkyPilot supports scalable model serving on Kubernetes and other providers. It uses a unified workflow to manage production ML operations, allowing you to serve models reliably alongside training and batch processing tasks.

How do I configure checkpointing and autostop for preempted ML jobs?

You configure checkpointing and autostop within the SkyPilot task YAML. This ensures preempted ML jobs survive interruptions by saving progress and automatically stopping idle clusters to optimize costs.

Does multi-cloud orchestration support automatic cluster selection for A100 GPUs?

Automatic cluster selection supports A100 GPUs by evaluating cost and availability across clouds. It identifies the best cloud, region, and accelerator combination for your workload constraints without manual intervention.