Fly.io

Deploys Docker containers as hardware-virtualized Fly Machines with dedicated CPU, memory, networking, and private filesystems. Each machine is isolated at the hypervisor level (not container-level), enabling untrusted code execution with guaranteed resource boundaries. Machines launch in under 1 second and consume resources only during active execution, with per-second billing for CPU and memory consumption.

Automatically distributes containerized applications across Fly's global infrastructure spanning 30+ geographic regions (Sydney, São Paulo, and others). Uses Anycast routing and edge-optimized networking to direct user traffic to the nearest regional instance, achieving sub-100ms response times. Developers specify deployment regions via configuration; Fly handles DNS resolution, load balancing, and traffic steering transparently.

Integrates with Elixir FLAME (Fly's distributed computing framework) to enable distributed task execution across multiple Fly Machines. Allows Elixir applications to spawn remote tasks on other machines and coordinate execution. FLAME handles machine provisioning, task scheduling, and inter-machine communication transparently.

Integrates with CockroachDB and globally-distributed Postgres to provide multi-region database support for Fly applications. Enables applications to read and write data with low latency across regions while maintaining consistency. Database instances can be deployed on Fly or external providers; Fly handles networking and connectivity.

Provides SSO integration for Fly.io account access and API authentication via narrowly-scoped tokens. Tokens can be restricted to specific organizations, applications, or operations, enabling fine-grained access control for CI/CD systems, third-party tools, and team members. Specific SSO providers and token scoping options not detailed.

Fly's infrastructure is built on memory-safe Rust and Go, reducing vulnerability surface from memory corruption bugs. This architectural choice affects platform reliability and security but does not directly expose capabilities to end users. Mentioned as security differentiator but implementation details not provided.

Attaches persistent block storage (NVMe) to Fly Machines for low-latency local data access, and provides global object storage for durable, replicated data. NVMe volumes are fast but ephemeral per-machine; object storage is slower but persists across machine restarts and regional failures. Developers mount volumes via fly.toml configuration and access object storage via standard S3-compatible APIs.

Provides built-in private networking allowing Fly Machines to communicate securely without exposing services to the public internet. Uses granular routing rules and end-to-end encryption (specific encryption standard not documented) to isolate traffic between machines. Machines are assigned private IPv6 addresses and can reference each other by internal DNS names (e.g., 'service.internal'). No additional VPN or networking infrastructure required.

Charges for CPU and memory consumption on a per-second basis, with no minimum commitment or always-on costs. Machines consume resources only during active execution; idle machines incur no charges. Developers can reduce costs by 40% by purchasing reserved capacity upfront, trading flexibility for predictable pricing. Pricing calculator available but specific per-second rates not disclosed in documentation.

Allows developers to capture the entire state of a running machine (memory, filesystem, processes) as a checkpoint, then restore from that checkpoint to resume execution or debug issues. Enables 'break something and restore' workflows without rebuilding or redeploying. Checkpoint mechanism details (API format, storage location, performance characteristics) not documented.

Integrates with CI/CD systems (specific integrations not named in documentation) to automate deployment from version control. Includes unique 'Accidental Deployments Are on the House' refund policy that credits charges for unintended deployments triggered by CI/CD misconfiguration or accidents. Integration mechanism and supported CI/CD platforms not detailed.

Command-line interface for managing Fly applications, machines, and deployments from the terminal. Enables developers to deploy, scale, monitor, and configure applications without using web UI. Specific CLI commands and capabilities not detailed in documentation. Supports declarative configuration via fly.toml files.

Provides GPU-equipped machines for accelerated AI inference and training workloads. GPU types and availability not detailed in documentation. Machines are provisioned via standard Fly configuration with GPU resource specification. Pricing and regional availability for GPU machines not documented.

Provides SOC2 Type 2 attestation for security and compliance requirements. Offers HIPAA-compliant infrastructure option for healthcare and regulated workloads, with Business Associate Agreements (BAAs) available. HIPAA compliance is an add-on service at $99/month. Specific security controls, audit logging, and data handling practices not detailed.