Fly

Fly.io API Skill

Quick Start

Fly.io transforms Docker containers into microVMs running globally. The Machines API allows you to programmatically spawn, pause, and destroy VMs in milliseconds.

  # Fly Machines API operates over standard HTTP REST

Common Workflows

Fast Machine Provisioning

Instead of scaling via traditional auto-scalers, use the Fly Machines API to launch a new microVM the moment a user requests a heavy workload (e.g., video rendering) and immediately destroy it once the workload completes.

Production Patterns

Global Anycast

Fly automatically routes HTTP traffic to the nearest microVM via Anycast. Ensure your database (e.g., Turso or a Fly Postgres read-replica) is also positioned in the same region to avoid cross-country latency penalties.

Error Recovery

Handle HTTP 503 errors if attempting to start a Machine in a region experiencing hardware capacity constraints. Always implement a fallback region (e.g., sjc falling back to lax).

Security Notes

Machines can communicate with each other over an encrypted private IPv6 network (6PN). Do not expose internal microservices to the public internet; restrict their services block in the fly.toml.

Performance Considerations

Machines boot incredibly fast (often <300ms), but your application runtime (e.g., a heavy JVM or massive Node.js payload) will dictate the actual cold start time. Optimize your Docker images aggressively.

Testing Guidance

You can interact with the Fly Machines API locally by utilizing fly proxy, which forwards traffic to your private Fly network without exposing it to the public internet.

Troubleshooting

If a Machine exits immediately after booting, check the logs (fly logs -m <machine_id>); this is almost always caused by an application crash (e.g., missing environment variables) rather than a Fly infrastructure issue.

References

• Machines API Docs

Related Skills

• Turso
• Render

Why use this skill

Use this when your agent works with fly — structured patterns beat pasted docs and prevent common hallucinations.

AI pitfalls

• Using outdated SDK or API versions from training data
• Inventing environment variable names
• Omitting error handling and retry logic

Production checklist

• Secrets in environment variables, not source code
• Error handling and logging in place
• Rate limits and timeouts configured

Related skills

• turso — related to
• render — related to

Last Verified: 2026-07-02