Abstract : Microblogs, although extremely peculiar pieces of data, constitute a very rich source of information, which has been widely exploited recently, thanks to the liberal access Twitter offers through its API. Nevertheless, computing relevant answers to general queries is still a very challenging task. We propose a new engine, the Twittering Machine, which evaluates SQL like queries on streams of tweets, using ranking techniques computed at query time. Our algorithm is real time, it produces streams of results which are refined progressively, adaptive, the queries continuously adapt to new trends, invasive, it interacts with Twitter by suggesting relevant users to follow, and query results to publish as tweets. Moreover it works in a decentralized environment, directly in the browser on the client side, making it easy to use, and server independent.