Social media have become an essential tool to collect timely information on news and events. Analyzing social streams in real-time for personalization and recommendation purpose have become important topics in the data management community. In this paper, we propose MYSTREAM, a personalization service to follow events from Twitter. To improve the scalability of the service, MYSTREAM adopts an in-browser and hybrid architecture. MYSTREAM leverages the device of users to perform the computational operations on the users' browser, while the server only provides all the necessary material to perform these tasks. In addition, MYSTREAM adopts a stream-based processing approach to identify the relevant contents in a real-time manner. Moreover, the recommendation engine of MYSTREAM is highly modular. Users can build a personalized dashboard by assembling the recommendation modules they prefer to follow the considered event. We implemented and evaluated MYSTREAM using real trace from Twitter. We show that MYSTREAM is effective to follow an event from Twitter, particularly the live recommendation module quickly identifies the most valuable contents over time. In a system perspective, we show that the cost running MYSTREAM on the client device remains minimal.