The asynchronous network assumption is a major research branch in Byzantine consensus protocols. However, despite being the weakest assumption among the current mainstream network models, it still differs significantly from real-world networks. While it assumes messages between honest nodes will eventually be delivered, this requires nodes to store and retransmit failed messages, leading to unbounded memory growth. Recent works such as Tusk (EuroSys 2022), Dumbo-NG (CCS 2022), and Honeybadger (CCS 2016) have proposed solutions to address memory limitations, but heavily rely on signatures. This not only introduces new cryptographic assumptions and setup for the underlying protocols but also incurs additional computational and communication overhead. To address this, we propose BFT-MS, a universal asynchronous BFT protocol compiler that bounds memory usage to a fixed, reasonable level without compromising security. BFT-MS operates in a near black-box manner, avoids cryptographic components, and adds no extra message exchanges beyond state transfer. Applied to Honeybadger and FIN, BFT-MS ensures memory usage stays below 2.5 GB even under adversarial attacks, while the original protocols exceed 32 GB within minutes. This demonstrates BFT-MS’s efficiency in managing memory without sacrificing performance.