Re:Boxxy puts bad Linux applications in a box with only thei...
by dloss on Wednesday August 28, 2024 @03:47AM
(#41377040)
Related question: Which software do you recommend to sandbox a locally running AI agent, so that it can only access parts of the filesystem (e.g. one folder) and an allow-list of URLs?
[ Parent
| on HN ]
Re:Boxxy puts bad Linux applications in a box with only thei...
by BonusPlay on Wednesday August 28, 2024 @04:14AM
(#41377215)
Depends how much patience you have.
Firejail if you want ease of use (there are a lot of ready profiles to be used).
Bubblewrap if you want more security, at the cost of having to do more manual work.
TL;DR Firejail is a blacklist of things, while bubblewrap is an whitelist, so bwrap policies tend to be tighter.
Re:Boxxy puts bad Linux applications in a box with only thei...
by OJFord on Wednesday August 28, 2024 @08:20AM
(#41378656)
That depends on the profile, firejail supports both.
Re:Boxxy puts bad Linux applications in a box with only thei...
by ravetcofx on Wednesday August 28, 2024 @03:52AM
(#41377072)
Chroot or AppArmor+Firejail
Re:Boxxy puts bad Linux applications in a box with only thei...
by amelius on Wednesday August 28, 2024 @05:14AM
(#41377575)
What are people using to quickly test new ML models from github in a sandbox?
Do you fire up a docker image? Do you use virtualbox? ...
Re:Boxxy puts bad Linux applications in a box with only thei...
by stavros on Wednesday August 28, 2024 @05:50AM
(#41377772)
What's allowing the AI agent to access files at all in the first place?