The choice and materialization of indexes are activities commonly done by database administrators to speed up database application processing. Due to the complexity of the index selection task and to the pressure for productivity increase put on tuning professionals, many works on the literature and on commercial systems seek for tools that can help the DBA choose the best indexes for a given workload. We classify these works as local self- tuning, once they are interested in a specific tuning problem, in opposition to global self-tuning work, which is targeted at obtaining acceptable performance for the system as a whole. This dissertation proposes two architectures that allow the complete automation of the index tuning task. Human intervention independence is achieved through the use of software agents. The combination of agents and DBMS makes systems more autonomous and self-tuning. We have implemented one of the proposed architectures in the open source DBMS PostgreSQL and obtained experimental results with a transactional workload that show the feasibility of our approach.