It uses OS level sandboxing. MacOS seatbelt, linux bubblewrap and windows restricted token. Agav also gives tool lvl permission and human approval for destructive actions. Sandboxing is applied at agent/tool exec rather than whole terminal session
Yes AI was used extensively to build this. I wouldn't call it vibecoded though, as changes from developers goes through review, tests and security checks much like any other good OSS project out there.
As for opencode its a very good OSS product, at agav we focus on two main verticals which are cost efficiency with minimal accuracy tradeoffs and secure execution through permissions, approvals and OS native sandboxing.
As for opencode its a very good OSS product, at agav we focus on two main verticals which are cost efficiency with minimal accuracy tradeoffs and secure execution through permissions, approvals and OS native sandboxing.