An approach for a biometric cryptosystem based on keystroke dynamics and the k-medoids algorithm is proposed. The stages that comprise the approach are training enrollment and user verification. The proposed approach is able to verify the identity of individuals offline avoiding the use of a centralized database. The approach as reported in this paper may be implemented in stand-alone terminals or embedded in password-based systems to increase the security. The performance of the proposed approach is assessed using 20 samples of keystroke dynamics from 20 different users. Simulation results show a false acceptance rate of 2.89% and a false rejection rate of 3.35%. The cryptographic key released by the proposed architecture may be used in several potential applications such as user login, file encryption or even portable authentication to gain access to virtual private networks.