CodeSandbox

Every sandbox runs inside an AWS Firecracker microVM with its own kernel and isolated file system. Snapshots of the full VM state are taken on idle and restored on resume, so reopening a project skips boot, dependency install, and build phases — typical resume time is around two seconds even for full-stack environments with running databases.

A single action forks the current microVM — including running processes, installed packages, open ports, and in-memory state — into a new independent VM. This enables parallel exploration of refactors, A/B testing of agent strategies, and lightweight 'what if' branches without re-bootstrapping the environment.

Node.js and Python SDKs expose microVM lifecycle (create, fork, snapshot, destroy), file system access, command execution, and port forwarding through a programmatic API. Designed for embedding inside AI agent backends, it supports persistent state across sessions and kernel-level isolation suitable for executing untrusted LLM-generated code.

Multiple users can edit the same files simultaneously with live cursors, share a single terminal session, and view forwarded ports together. Built-in to every sandbox without separate setup, making it suitable for pair programming, interviews, workshops, and live debugging sessions.

Developers can work in the browser-based web editor, the official VS Code extension, or the JetBrains Cloud Containers plugin — all backed by the same remote microVM. This lets individual contributors keep their preferred local editor while the team shares a consistent cloud environment.

Repositories sync bidirectionally with GitHub, and pull requests can automatically spin up ephemeral preview environments built from the same microVM snapshots, giving reviewers a working app to test rather than just a diff.