The latest release of agentscope-ai/qwenpaw is v2.2.2.post1. QwenPaw is a personal AI assistant harness that you deploy locally or in the cloud and connect to multiple chat apps. However, the provided FACTS contain no changelog for this version, so it is not possible to confirm what changed in this release.
The README explains extending functionality with skills and plugins, and describes a 3-tier memory structure divided into working context, conversation history, and knowledge base. The QwenPaw-Flash model list includes 2B, 4B, 9B, 35B-A3B, and 27B, and states support for local runtimes such as Ollama and LM Studio as well as more than 14 cloud providers. Its security features include kernel sandboxing, tool and file protection, and access policies. It is written in TypeScript and uses the Apache-2.0 license. This configuration is worth viewing as a design case for an assistant harness that lets you choose the deployment location, chat channels, and extensions.
The basis is the functionality listed in the repository description and the README; the provided FACTS contain no material verifying the performance or security effects of those features. The specific changes in v2.2.2.post1 were not given either. Therefore, the features and structure presented in the documentation can be confirmed, but it is difficult to judge the scope and stability of real-world behavior.