perplexityai/SPACE

What it solves

SPACE (Sandboxed Platform for Agentic Code Execution) provides a secure, isolated environment for running untrusted code. It is designed for applications—such as AI agents or code-running services—that need to execute user-provided or generated code without risking the host system's security or stability.

How it works

SPACE uses Firecracker microVMs to create lightweight, isolated Linux environments. Applications interact with the platform via an HTTP API to manage the entire lifecycle of a sandbox: creating environments from templates, executing commands, streaming output back to the application, and pausing or deleting the sandbox when finished. It supports advanced features like warm pools to reduce startup latency and snapshotting to restore environments quickly.

Who it’s for

This platform is for developers building AI agents, code interpreters, or collaborative coding platforms that require a robust, scalable, and secure way to execute arbitrary code in a controlled environment.

Highlights

• Uses Firecracker microVMs for high-performance, secure isolation.
• Provides a language-agnostic HTTP API for sandbox management.
• Supports warm pools to minimize cold-start times for agentic workflows.
• Includes an E2B compatibility wrapper to support existing SDKs.
• Built with a focus on observability and production-grade reliability.