I’m working on an OSS version of something very similar to this, but the angle is more sovereignty and security over AI agents. Will post on HN when it’s ready.