Your AI agent is writing code nobody reviewed and running it on your machine.
Most teams answer "where does it run" with a container and a hope.
Introducing CreateOS Sandbox.
Run code you cannot trust, with the networking, egress control, and self-hosting you need. π§΅
@mattpocockuk docker sandbox on WSL is a nightmare. We ran into the same issue and built a micro-VM sandbox instead.
Own kernel, egress locked down at the kernel level. No way for the agent to talk its way out. DM us if you want to try it
The news is out! CreateOS Sandbox is live.
A micro-VM for running untrusted code. Own kernel. Locked down egress. Pause when idle.
Follow the link below and spin up your first Sandbox.
Your AI agent is writing code nobody reviewed and running it on your machine.
Most teams answer "where does it run" with a container and a hope.
Introducing CreateOS Sandbox.
Run code you cannot trust, with the networking, egress control, and self-hosting you need. π§΅
@haha_girrrl At CreateOS, we see it a little differently.
The goal is shipping value. We focus on providing the infrastructure that enables builders to prototype their vision and scale their substance π
Spin up a full Linux micro-VM in seconds. Own kernel, own filesystem, own network. Pick a size, hit create, and you are in a root shell. No SSH key needed.
Use it to run AI-generated code safely, spin up dev boxes, or run CI that tears down when done.
Walkthrough ππ»
This is what real isolation for AI agents looks like: a full Linux VM, not a container with extra steps. Own kernel, own filesystem, own network, booting in seconds.
Watch the full explanation below by @pratikbin ππ»π¬
Most sandboxes for AI agents are containers with extra steps: same kernel as your host, same blast radius.
CreateOS Sandbox is a full Linux VM: own kernel, filesystem, network. Firecracker microVMs, boots in seconds.
Each sandbox is a Firecracker microVM. The isolation of a full VM, the boot speed of a container. Own kernel, own root filesystem, own network identity. Ready to run commands within seconds of creation.
Built for four things: running untrusted or AI-generated code without risking your own machine, ephemeral dev environments per branch or per task, CI and batch jobs that tear down when finished, interactive sessions over SSH or HTTPS.
Pause a running sandbox to durable storage. Resume it later. Fork it into an independent copy. Run commands and stream output over the API, SDK, or CLI. No SSH key required.
Three ways in, one control plane: REST API, TypeScript SDK, CLI. This week's demo covered the dashboard, the CLI, and the SDK: create a sandbox, run two commands, tear it down. About 35 lines of code end to end, comments included.
Networking and storage: private networks so sandboxes reach each other by name, isolated from other tenants. Mount S3-compatible disks live. Expose a public HTTPS URL per sandbox, or lock egress to an allowlist.
Billing is per second. You pay for the machine while it is alive. Paused sandboxes cost nothing.
What would you run in a disposable Linux VM that spins up in seconds?
Try CreateOS Sandbox at: https://t.co/Grujqfumc9
Your AI agent is writing code nobody reviewed and running it on your machine.
Most teams answer "where does it run" with a container and a hope.
Introducing CreateOS Sandbox.
Run code you cannot trust, with the networking, egress control, and self-hosting you need. π§΅
@naman_307 Spot on.
CreateOS is putting the power back in the hands of the enterprise and building that foundation, and the sandbox is one massive, foundational leap in that direction.