E2B

E2B provides secure sandboxes for AI agents: short-lived cloud computers where an agent can run the code it writes, install packages, work with files, and use the network without touching the developer's own infrastructure. It is the execution layer behind code-interpreter features, data-analysis agents, coding agents, and agent evaluation pipelines, and it is one of the built-in sandbox providers in the OpenAI Agents SDK.

Each sandbox runs in its own Firecracker microVM, the virtualization technology behind AWS Lambda, so every session gets a dedicated kernel. That is a stronger isolation boundary than shared-kernel Docker containers, which matters when the code comes from a model rather than a trusted developer. Sandboxes start in roughly 150 to 200 milliseconds from prebuilt templates, can be paused and resumed with their state intact, and run for up to an hour on the free plan or up to 24 hours on Pro.

Developers control sandboxes through Python and JavaScript SDKs. The Code Interpreter SDK adds Jupyter-style execution that streams stdout, errors, and generated charts back to the application. Custom templates bake in dependencies, a Desktop Sandbox provides a full graphical Linux desktop for computer-use agents, and an MCP gateway built with Docker gives agents inside a sandbox access to 200+ tools from the Docker MCP Catalog.

The SDKs and the infrastructure are Apache-2.0 open source. The managed E2B Cloud bills compute per second, with a free Hobby plan and a Pro plan for longer sessions and higher concurrency. Teams that need sandboxes inside their own cloud can self-host the infrastructure with Terraform and Nomad on Google Cloud, with AWS support in beta, which takes real operational effort. The company is based in Prague and San Francisco, and its funding includes a $21M Series A.