Alibaba releases AgentScope 2.0: Addressing environmental decoupling and security defense issues during Agent deployment According to Beating monitoring, Alibaba Tongyi Laboratory has released AgentScope 2.0, a multi-agent development framework. Compared to the "transparent development" positioning of 1.0, which focuses on visual display of messages, the new version focuses on high availability and security control of intelligent agents in real production environments, attempting to solve pain points such as long task interruptions and permission loss. At the call and message layers, the framework introduces retry and backup model mechanisms to prevent task interruption caused by a single model timeout. Refactoring the message module into a Content Block that supports multimodal streaming data and provides streaming output and human in the loop capabilities in conjunction with event systems. In response to the security risks of Agent autonomous operation, 2.0 has added a permission system that can statically and dynamically intercept high-risk shell commands and sensitive file reads and writes. In addition, 2.0 introduced Workspace environment abstraction, decoupling the business logic of agents from the execution environment. The same intelligent agent can run locally, in Docker containers, or in E2B cloud sandboxes without modifying the code; Combined with the preheating pool mechanism, it can reduce the container creation overhead in high concurrency scenarios such as reinforcement learning. At present, the framework has completed the 2.0 upgrade of the Python version and has launched a new TypeScript version. The Java version will also be updated in the near future. [Original link]