Twitter is one of the most used social networks and its specific features makes it well suited for the real-time analysis of geographic trends of a specific topic. Earle et al. (2011) have shown how the analysis of Twitter streams can provide a useful tool for the early detection of earthquakes at a global scale. They proved that data mining of social networks could provide useful information in Seismology. Here we present a software system named TwiFelt, aimed at providing real-time earthquake perception maps from the analysis of Twitter streams. The system is based on the collection of geotagged tweets (i.e. tweets having a geographic reference) containing selected keywords, its statistical interpretation and its interactive graphical representation. Figure 1 represents a schematic overview of the system. The most important parts of TwiFelt are: a Twitter stream parser (see sec.1), a database (see sec.2) and an interactive web interface (sec.4). The web interface provides maps showing, using a shaded overlay, an indication of the area where an earthquake has been felt. In the following we describe the system components, showing also some preliminary results. Since most of the references to the mentioned technologies are currently only available online, the relevant links are listed in a separate Appendix.