What it does
Hardware-virtualized containers. Per-second billing. Global placement (the user picks regions; Fly routes requests). The base compute primitive the rest of the Fly product line is built on.
Target ICP
Framework-shipping developers (Ring 1). Anyone with a Dockerfile or a supported framework can fly launch and be running in minutes.
Inputs / Outputs
- Input: Docker image (or framework-scaffolded app via
fly launch) + region spec + resource spec (CPU, RAM). - Output: Running VM with public URL, TLS, private networking, optional volume, optional Postgres.
How it works
Firecracker-based microVMs. The user specifies size (shared CPU, 256MB–8GB RAM; dedicated CPU up to much larger). Per-second billing starts on fly deploy, stops on scale-to-zero. Can be configured for auto-start/auto-stop on request.
Pricing / limits
Metered pay-as-you-go. 40% discount with Machine Reservations. Free allotment referenced but scope unclear. See Flydotio - Pricing.
UX entry points
- CLI:
fly launch→fly deploy. - Docs: top-level "Fly Machines" section.
- Machines API — programmatic surface used underneath flyctl.
Differentiators (claimed)
- Hardware-level isolation via microVM rather than shared-kernel containers.
- Per-second billing vs. minute or hour billing at competitors.
- Global-by-default — regions are a first-class deploy argument.
Integrations
- Postgres, Tigris, Upstash referenced as first-class add-ons in docs.
- Kubernetes via Fly Kubernetes.
Sources
- fly.io/docs — Fly Machines section