It’s interesting that they’re using Blitz for parsing HTML/CSS, which is at an early stage with a lot of features still not implemented, which doesn’t matter for their use-case of AI agent-only browsing.