Mainbrella

Watch Mainbrella demo

About Mainbrella

Mainbrella provides isolated, secure Linux workspaces specifically designed to host AI agents, evals, and automated developer workloads. It deploys sandboxed microVMs on demand, enabling developers to run shell commands, install software packages, edit files, and transfer assets programmatically via a unified HTTP API. Every workspace runs in its own kernel with network isolation, ensuring that agent execution remains secure and entirely separate from the host application.

Who is Mainbrella for?

AI engineers and developer teams who previously had to run untrusted agent-generated code directly on their own local machines or manage complex, costly, and insecure custom Docker clusters. They struggled with safety risks when agents executed arbitrary commands, and faced difficulties maintaining clean environment states across multiple runs.

How does Mainbrella work?

First, spin up an isolated container by calling the creation endpoint with your preferred runtime image. Next, execute bash commands, edit files, or run test suites programmatically using the HTTP API. Finally, save your progress to preserve the filesystem for future runs, or let the workspace cleanly shut down to discard temporary changes.