Deep Learning has been shown to be very effective when applied to binary analysis. In this talk we will present SAFE, an open source tool using a recurrent neural network to create similarity preserving vectorial representations of binary functions (the so called embeddings). Using them it is possible to detect if two functions are similar or not, thus embeddings can be a useful support for reverse engineering. As example, they permit to identify library functions inside a binary even if we don't have access to the exact compiled version of the linked library. Moreover, embeddings can be used also as a signature for vulnerability discovery and malware hunting. Finally, embeddings permit to identify the behaviour of a function, such as encryption.