The roles seems more oriented towards visual agents. Is this is similar to google vision agent but with more capabilities?