E2B

E2B gives AI agents a computer. Each sandbox is an isolated cloud VM with a filesystem, a shell, internet access, and a runtime, started in about 150 milliseconds and destroyed when the task ends.

Why it exists

Any agent that writes code eventually needs to run it, and running model-generated code on your own infrastructure is a bad idea. E2B provides the isolation boundary, so a mistake — or a prompt injection — is contained to a disposable VM.

Capabilities

• Fast start-up, fast enough to create a sandbox per request rather than pooling them
• Filesystem access for uploading data and retrieving generated artefacts
• Process control for running commands and streaming output back
• Code interpreter SDK with a Jupyter-style kernel, stateful across calls, returning charts and tables as structured results
• Custom templates so a sandbox boots with your dependencies preinstalled
• Long-running sandboxes with pause and resume for multi-hour tasks

Who uses it

Teams building data analysis agents, coding agents, AI app builders, and evaluation harnesses. SDKs for Python and JavaScript; open source with a free tier, usage-based pricing by sandbox seconds, and self-hosting for enterprises.