The Practical Bizantine Fault Tolerance (PBFT) consensus is a well adopted consensus protocol in private and consortium blockchains, the reason being that it is a light protocol, it is not computational intensive (particularly when compared to proof-based consensus protocols), and it allows to reach high transaction throughput. Security in this consensus depends on the number of nodes (aka validators) involved in the consensus: the higher the number of nodes, the higher the security level. However, increasing the number of validators goes along with an increase in the latency, and consequently a decrease in the throughput. The present research proposes a new Self- Adaptive BlockchaIn coNsEnsus protocol (Sabine), which tries to optimally address this trade-off by continuously adapting the size of the pool of validators with the prime objective that the output throughput (i.e., number of transactions validated at a given time) meets – to the best extent possible – the input throughput (i.e., requested by the application). Sabine is evaluated and validated in real-life settings, whose results show that Sabine has a relative error of 7.79% between the requested and committed transaction throughput, compared to 39.5% for a classic chain with all nodes participating in the consensus, implying a more reactive chain, with less latency.