The Tencent WeChat team has made WeKnora publicly available—a corporate knowledge platform that transforms documents into a RAG search stack, a ReAct-style agent runtime environment, and a self-sustaining Wiki.
The Tencent/WeKnora repository on GitHub has garnered over 25,000 stars, making this project one of the most notable Chinese-origin agent knowledge bases for developers. These developers require a system that provides information retrieval, tool usage, and long-term memory within a single solution, rather than just a chat wrapper.
The latest versions of WeKnora significantly surpass classic RAG. Materials from version 0.8.0 describe a skills sandbox with Docker, E2B, and Cube backends, a catalog of skills for tenants, long-term cross-session memory covering profiles, preferences, facts, tasks, and interests, support for knowledge graphs, MCP-oriented tools, and Wiki-style document servicing. This system is designed to allow agents to search corporate corpora, invoke skills according to network policy, and maintain persistent memory between sessions, which is closer to an agent's operational level than a bot for single-query responses.
Official documentation also points to the WeChat Dialog Open Platform paths for Official Accounts and Mini Programs if teams need to utilize the same knowledge base on messaging surfaces.
The integration mechanism crucial for coding agents is the official DeepSeek Harness plugin, published as @wxg-prc-cpg/dsh-weknora. DeepSeek Harness itself provides workspace search and web search but lacks a built-in embedding path or knowledge base; this plugin fills that gap by registering four read-only tools—knowledge base listing, hybrid search, full document reading with page scrolling, and WeKnora-generated answers with citations—against a deployed instance of WeKnora.
Installation is documented via the command dsh plugin --profile web add @wxg-prc-cpg/dsh-weknora, which was tested on dsh 0.1.0-rc.8 and WeKnora 0.7.2, with credentials passed via an API key, and no write paths to the corpus.
For teams already evaluating DeepSeek Harness for agent workflows, WeKnora represents the missing piece of corporate information retrieval: documents are stored in WeKnora, and the harness agent calls them as tools, leaving indexing and chunk management to WeKnora. This focus distinguishes this story from simple DeepSeek Harness feature announcements. Operators should still independently verify sandbox isolation, API key handling, and citation quality on their own corpora before considering such a combination production-ready.