WeChat AI team has officially open-sourced the knowledge management framework WeKnora (Vinerah). The latest 0.8.0 version is now fully available to the public. Its positioning can be summarized in one sentence: making knowledge bases not just "able to find answers," but "able to be executed."
WeKnora aims to solve the most practical problem when large models are deployed—knowledge lies in documents, the model speaks fluently, but turning them into verifiable results requires a pair of hands and feet in between.
anydoc parsing + GraphRAG, transforming messy documents into knowledge graphs
The first highlight is anydoc parsing: various formats such as PDF, Word, Markdown, and web pages can be directly imported without prior conversion; combined with Wiki mode, it can automatically organize messy documents into structured knowledge systems.
The second is GraphRAG. While traditional RAG retrieves relevant fragments via vector search and presents them to the model, GraphRAG goes a step further—extracting entities and relationships to build a graph. This makes it much more reliable when answering questions about how things are related.
Skill Sandbox Runtime: Knowledge Can Be Safely "Executed"
The most notable feature is Skill Sandbox Runtime. Content from the knowledge base can be encapsulated into "skills" and run in sandbox environments like Docker, E2B, CubeSandbox—meaning knowledge is not just "read," but can be safely "executed." The model can call a skill from the knowledge base, run it in the sandbox, and return a verifiable result, rather than stopping at "I suggest you do this."
Combined with the long-term memory (Long-term Memory) function, this framework has a very complete intention: parsing, organizing, retrieving, executing, and memorizing form a complete workflow. It adopts a modular architecture, allowing users to select what they need; the license is MIT, which is open-source friendly and has no commercial burden, which is also the confidence for it to face scenarios from individuals to enterprises.
