chrisscho.uk
[ CASE FILE // developer labs ]Active

open-desktop

Cloud desktop infrastructure and container sandbox engine for autonomous AI agents.

A sandboxing layer that gives an autonomous agent a disposable desktop and container to work in, so it can be handed real tools without being handed the host machine.

An agent that can only read files is safe and not very useful. An agent with a shell on your laptop is useful and not very safe. open-desktop is the middle: a throwaway containerised desktop an agent can be given, break, and be handed a fresh copy of. It is the public one of my research repos, so the source is there to read.

Where it fits

Research track. The question underneath it is how much autonomy you can safely give an agent, which is the same question every client build eventually runs into.

Stack
PythonDockerLinux
[ ILLUSTRATION ]
$ open-desktop spawn --profile agent-sandbox
[14:20:03] CONTAINER: provisioned sandbox (cpu=2, mem=4G, net=restricted)
[14:20:05] DESKTOP: virtual display :99 up, session recording on
[14:20:06] AGENT: handed control, host filesystem not mounted

Want a system like this built for your business? See consulting services →