Vehicle-sharing systems (VSSs) have become massive, requiring increasingly efficient management strategies. A critical aspect for properly operating these systems is managing imbalance, especially for free-floating fleet schemes where users pick up and release vehicles anywhere in the geographical area of influence. For these types of imbalance problems, strategies are needed to detect early situations of vehicle crowding in certain areas and redistribute them appropriately. This paper presents a multi-agent model for free-floating VSSs that divides the geographical area into zones and assigns an agent to analyse and predict the local situation based on traffic and incomplete demand information. Then, this local information is sent to a central agent in charge of calculating the possible redistribution of vehicles to minimise the impact of the imbalance. The simulations performed on 24 h of bike-sharing traffic in the city of Tartu in Estonia show that, by using the proposed algorithms, a significant reduction of the imbalance is obtained.