Cloud computing is an opened and distributed network that guarantees access to a large amount of data and IT infrastructure at several levels (software, hardware...). With the increase demand, handling clients’ needs is getting increasingly challenging. Responding to all requesting clients could lead to security breaches, and since it is the provider’s responsibility to secure not only the offered cloud services but also the data, it is important to ensure clients reliability. Although filtering clients in the cloud is not so common, it is required to assure cloud safety. In this paper, by implementing multi agent systems in the cloud to handle interactions for the providers, trust is introduced at agent level to filtrate the clients asking for services by using Particle Swarm Optimization and acquaintance knowledge to determine malicious and untrustworthy clients. The selection depends on previous knowledge and overall rating of trusted peers. The conducted experiments show that the model outputs relevant results, and even with a small number of peers, the framework is able to converge to the best solution. The model presented in this paper is a part of ongoing work to adapt interactions in the cloud.