Sprites are persistent, hardware-isolated Linux execution environments designed for arbitrary code. Developed by Fly.io, they function as small, stateful computers that can be spun up on demand, offering a simpler alternative to managing complex infrastructure. Unlike traditional serverless functions, Sprites maintain their entire filesystem and memory state between runs, providing a consistent environment for various workloads. They are built on Firecracker VMs, ensuring strong isolation and secure execution.

Sprites dynamically manage underlying resources, billing users only for active compute usage, with no charges incurred while the environment is idle. This model combines the benefits of persistent virtual machines with the cost-efficiency of serverless platforms. The platform provides a CLI, REST API, and SDKs for programmatic control, allowing seamless integration into workflows and applications. Sprite environments come pre-configured with a broad suite of development tools and languages, making them ready for immediate use across a range of programming tasks.

Some of the key features are:

• Hardware-isolated Execution Environment: Code runs within dedicated Firecracker VMs, providing robust security and isolation.
• Persistent ext4 Filesystem: All files, installed packages, Git repositories, and databases are preserved between runs on a standard ext4 filesystem. Data is stored on fast NVMe storage during active use and backed up to durable object storage when idle.
• Automatic Idle Behavior: Sprites automatically transition to a warm state (VM suspended, no compute billing, quick wake-up) and potentially to a cold state (in-memory state dropped, slightly longer wake-up) when inactive.
• On-demand Wake-up: Environments wake up automatically within 100-500ms (warm) or 1-2s (cold) upon command execution or an incoming HTTP request.
• Unlimited Checkpoints: Users can create transactional point-in-time snapshots of the entire Sprite filesystem, including files, installed packages, databases, permissions, and running processes (which stop during checkpointing). Checkpoint creation takes approximately 300ms.
• Quick Restores: Environments can be quickly restored to any previous checkpoint, replacing the current filesystem state and restarting the Sprite.
• HTTP Access with Unique URLs: Each Sprite is assigned a unique URL (e.g., https://<name>.sprites.app) that can be configured for public or authenticated access, routing inbound HTTP traffic to port 8080 by default.
• Dynamic Resources: Sprite VMs can access up to 8 CPUs and 16GB of RAM, consuming only the resources necessary for the current workload.
• Tiered Storage: Data is cached on fast, directly attached NVMe storage and backed up to durable external object storage. Each Sprite starts with a 100GB partition that can scale as needed, with billing based only on actual data written.
• Granular Billing: Billing is based on actual CPU cycles, resident memory, and storage consumed, measured hourly with per-second granularity for compute resources.
• Network Policy: Fine-grained Layer 3 network policies restrict outbound connections to specified domain names, configurable via API with immediate enforcement.
• Pre-installed Tools: Environments include popular languages such as Node.js, Python, Go, Ruby, Rust, Elixir, Java, Bun, and Deno, along with essential development utilities like Git, curl, wget, vim, and AI/CLI tools like Claude CLI and OpenAI Codex.
• Sessions: Supports detachable TTY sessions, enabling users to start long-running processes, disconnect, and reattach later.
• Services: Provides a mechanism for managing persistent background services that auto-restart whenever the Sprite wakes up, ensuring web servers or other background tasks are ready to handle requests.
• Port Forwarding: Allows tunneling TCP connections from a local machine to specific ports within a Sprite, useful for database access or debugging.
• Tasks API: Enables long-lived processes or AI agents to register tasks that keep a Sprite in an active state, preventing it from pausing and dropping open TCP connections.

Sprites operate by providing a shell-like experience where users interact through a CLI or SDKs. After initial setup and authentication with a Fly.io account, a user can create a new Sprite. Once active, commands can be executed or an interactive console session can be opened. The system automatically handles the lifecycle, suspending the Sprite when idle to save resources and waking it rapidly upon the next interaction, whether it's a command, API call, or an HTTP request to its dedicated URL. Checkpoints can be created to save the environment's state before making significant changes and restored if needed.

Some common use cases include:

• AI Code Execution: Running code generated by large language models (like Claude Code) in a secure, isolated sandbox.
• Untrusted Code: Safely executing user-submitted code or binaries without compromising the underlying system's integrity.
• Development Environments: Creating persistent and reproducible development environments that retain installed packages, files, and configurations across sessions.
• Long-lived Services: Hosting web services, APIs, or background processes that can automatically go idle when not in use and resume instantly on demand.
• CI/CD Tasks: Utilizing Sprites as ephemeral yet stateful environments for continuous integration and continuous delivery (CI/CD) pipelines, enabling quick testing against live repositories with full environment access.