AI News

WeChat’s AI team has open-sourced WeKnora, a knowledge-management framework whose public 0.8.0 release is intended to let knowledge bases do more than retrieve answers. It combines anydoc parsing with a Wiki mode that can import PDFs, Word files, Markdown, and web pages without prior format conversion, then organize unstructured material into a structured knowledge system. WeKnora also supports GraphRAG, which extracts entities and relationships into a graph to address questions about how pieces of information are connected. Its most distinctive component is Skill Sandbox Runtime: knowledge-base content can be packaged as skills and run in Docker, E2B, or CubeSandbox environments. A model can call a skill, execute it in a sandbox, and return a verifiable result rather than only offering procedural advice. The framework also includes long-term memory, linking parsing, organization, retrieval, execution, and memory in one workflow. Its modular architecture allows components to be used as needed, and its MIT license permits commercial use. The article presents it as suitable for scenarios ranging from individuals to enterprises.