Why every agent gets its own virtual machine
Agents that run code and browse the web need real computers. Here is how we give them one without risking yours.
An agent that can only talk is limited. The useful ones run code, download files and use websites, which is exactly what makes them risky to run next to your data.
A micro-VM per agent
In vvek every agent works inside its own Kata Containers micro-VM, with its own Linux kernel. Inside, the agent has root, a 10 GB home folder that persists between turns, and a browser on a virtual screen. Outside, the VM can reach the public internet but nothing inside our platform.
Recorded, so you can check
Every command the agent runs and every page it opens is recorded: the output, a screenshot, a video and the list of requests. Open a ticket's Work tab to replay it, or watch the agent's screen live.
Cheap where it can be
Routine work such as reading a screenshot or transcribing a voice note runs on our own hosted models, so it costs a fraction of a frontier model. When you want a frontier model, bring your own key.
