In stream processing, managing sensitive information in a timely manner while ensuring trust remains a significant challenge. When parties without a priori trust cooperate to execute a streaming application, it is difficult to ensure that sensitive data is kept confidential while guaranteeing that every party executes their code honestly.

In recent years, part of the web is shifting to a client-centric, decentralized model where web clients become the leading execution environment for application logic and data storage. However, current solutions to build decentralized web applications with multiple distrusting parties often involve a decentralized backend of servers running a BFT protocol between them. Existing consensus protocols using either all-to-all communication, or leader-based gossip suffer from performance degradation in unstable network conditions. In this paper, we present BeauForT, a purely browser-based platform for decentralized BFT consensus in client-centric, community-driven applications. We propose a novel, optimistic, leaderless, gossip-based consensus protocol, tolerating Byzantine replicas, combined with a robust and efficient state-based synchronization protocol. This protocol makes BeauForT well suited for the decentralized client-centric web and its dynamic nature with many network disruptions or node failures.