What are Fly Machines?

Fly Machines are lightweight virtual machines on Fly.io that can be created, configured, started, and stopped programmatically via a REST API. Unlike traditional VMs that take minutes to provision, Fly Machines boot in seconds and can be managed entirely through API calls -- making them ideal for platforms that need to spin up isolated compute instances for each customer.

Fly Machines run on Firecracker, the same microVM technology that powers AWS Lambda. Each machine gets its own kernel, filesystem, and network stack, providing strong isolation between tenants. But unlike Lambda, machines can run any Docker container, persist data on attached volumes, and stay running indefinitely.

For AI agent hosting, Fly Machines solve the per-tenant isolation problem elegantly. Each customer gets a dedicated machine running their agent, with complete filesystem separation, independent resource limits, and a unique network identity. The platform can provision, configure, and manage thousands of these machines through the API.

How Fly Machines Work

API-driven lifecycle -- Create, start, stop, restart, and destroy machines through REST API calls
Fast boot -- Machines start in seconds, enabling rapid provisioning of new tenant environments
Docker-based -- Run any Docker image without modification, including complex applications like OpenClaw
Persistent volumes -- Attach durable storage for data that survives machine restarts
Global regions -- Deploy in 30+ regions worldwide for low-latency access and data residency compliance
Private networking -- Machines in the same organization can communicate over a private WireGuard mesh network

Why Fly Machines Matter

Fly Machines fill a gap between serverless functions (ephemeral, limited) and traditional VMs (slow to provision, over-provisioned). They are fast enough for on-demand provisioning, persistent enough for stateful workloads, and isolated enough for multi-tenant platforms. The API-first design makes them programmable infrastructure rather than click-to-provision servers.

How KiwiClaw Uses Fly Machines

KiwiClaw provisions a dedicated Fly Machine for each tenant during onboarding. The orchestrator service creates the machine via the Fly Machines API, injects tenant-specific configuration through environment variables, and attaches a persistent volume for agent data. Each machine runs a stock OpenClaw Docker image. The LLM proxy runs as a separate Fly app accessible over private networking, and tenant machines get unique subdomains (slug.kiwiclaw.app) with automatic TLS certificates.

Related Terms

Frequently Asked Questions