How WeKnora Turns Enterprise Documents Into an Agent-Ready Knowledge Layer

WeKnora is an open-source knowledge platform from Tencent’s Weixin team. Its central idea is to make the same organizational documents useful in three ways: quick retrieval-augmented answers, multi-step agent work, and a Wiki that organizes source material into linked pages. The DeepSeek Harness integration adds another way for an agent to access that knowledge; it is not a replacement for WeKnora itself. 11621

One knowledge base, three ways to use it

RAG Quick Q&A retrieves material from a knowledge base to answer everyday questions. ReAct Agent is intended for tasks that need several steps: it can retrieve documents, search the web, and call MCP tools and tenant skills. Wiki Mode distills documents into interlinked Markdown pages with a visual knowledge graph. The project also describes manual editing, revision history, and rollback for those pages. “Self-maintaining” describes the intended workflow, not a reason to stop checking generated pages against their sources. 18

The newer agent capabilities extend beyond search. WeKnora describes session-persistent skill sandboxes with Docker, E2B, or Cube backends for working with files and running tasks. Its cross-session memory can retain useful context; Weixin says users approve suggested memories before they are used. WeKnora can also expose a workspace through an MCP endpoint so external clients can search and ask questions. 161216

WeKnora is a Weixin-team project, and reporting links it to the WeChat Dialog Open Platform. Separately, descriptions of its channels include enterprise WeChat and other messaging integrations. Those connections should not be confused with the DeepSeek Harness plugin, which is a distinct route into a deployed WeKnora service. 451621

What the 25,000-plus GitHub stars mean

A September review reported 25,863 stars for the WeKnora repository, while an earlier September report cited 23,400. These are snapshots, not a fixed current count. They indicate substantial developer attention, but cannot establish retrieval accuracy, access-control behavior, security, or reliability in a particular deployment. 46

How the DeepSeek Harness plugin works

The @wxg-prc-cpg/dsh-weknora plugin gives a Harness agent four read-only tools for a deployed WeKnora instance: list accessible knowledge bases, run hybrid search, read a document in pages, and request a WeKnora-generated RAG or agent answer with citations. The cited-answer tool can return a session ID for follow-up requests. “Read-only” describes the tools exposed to Harness; it does not certify every internal action of a WeKnora agent workflow as side-effect-free. 2124

The division of labor matters: WeKnora supplies the document corpus, retrieval, and any answer it generates; Harness orchestrates its own agent and decides how to use the returned passages or answer. WeKnora’s HTTP API documents workspace-level and platform-level API keys, as well as other authentication options, so the credential and its scope need to match the deployment and plugin configuration. A Harness model credential serves a separate purpose. 172136

For a Harness web profile, the published package-install command is: 21

dsh plugin --profile web add @wxg-prc-cpg/dsh-weknora

This installs a plugin, not a WeKnora server. Point it at an already deployed, reachable instance and follow the package instructions for its configuration and credentials. The package listing shows WEKNORA_BASE_URL as a deployment setting; consult its full instructions rather than guessing other environment-variable names. 21

Versions and production checks

September reports say the integration was verified with DeepSeek Harness 0.1.0-rc.8 and WeKnora 0.7.2. A plugin directory separately lists dsh-weknora 0.1.0 as “Verified,” while WeKnora documentation describes v0.8.0 capabilities. Those statements refer to different components and do not establish that every newer combination has been tested together—or that a directory badge is a security review. 120424348

Before using the integration with sensitive documents, pin and test the actual server, plugin, and Harness versions together. Check which knowledge bases and documents the configured credential can access; verify citation quality, full-document pagination, timeout handling, and what confidential text reaches the calling agent. GitHub popularity and a plugin listing are useful discovery signals, not substitutes for those deployment checks. 243648