Tencent open-sources WeKnora: one self-hosted knowledge base for RAG, agents and wiki pages

According to a post on dev.to, Tencent has released WeKnora, an open-source project aimed at a familiar workplace problem: teams rarely lack information, but it is fragmented across files, folders and services, so finding an answer means hunting through them first. WeKnora ingests the documents a team already has and turns them into a single knowledge system that can be queried in three different ways.

Three interfaces to one knowledge base

The first interface is a retrieval-augmented generation mode that returns quick answers grounded in the source documents, with citations attached so readers can verify where each claim came from. The second is a multi-step agent that can call tools and apply skills, which suits questions needing more than a single lookup. The third is a wiki: the same knowledge, organised into linked pages that are assembled from the underlying documents rather than written and maintained by hand.

Because all three interfaces sit on top of the same corpus, there is less of the drift that occurs when a chat assistant, an automation agent and a company handbook each keep their own version of the truth.

Keeping pages consistent with their sources

The dev.to write-up highlights a mechanism designed to keep the wiki aligned with its source material. Uploading a new document merges its content into pages that already exist, rather than spawning duplicates. Deleting a document goes further: the pages that drew on it are rewritten, or removed entirely if nothing remains to support them.

This targets a chronic weakness of both hand-maintained wikis and RAG pipelines, namely staleness. A knowledge base that quietly retains outdated material can be worse than an empty one, because the old content still looks authoritative.

Version 0.9.0 and adoption

Version 0.9.0 shipped on 29 October 2026 and added three capabilities: an assistant built into the application itself, collaboration between multiple agents, and support for automations. The dev.to post also reports that the project's GitHub repository has accumulated 36,779 stars, a figure that points to substantial interest, though it reflects the moment the post was written.

Deployment options

WeKnora can be run in two ways: as a Docker-based self-hosted deployment, or as a single "Lite" binary. Both approaches keep the documents, and the queries run against them, on infrastructure the team controls. That matters for internal documentation and other sensitive material that organisations would rather not hand to a third-party service.

Why it matters

Teams wanting this capability today typically assemble it from parts: a vector store, an agent framework, an orchestration layer, and a wiki that nobody keeps updated. WeKnora collapses those pieces into one open-source, self-hosted package in which the wiki is generated and maintained from the documents themselves.

The deletion behaviour is the standout detail. RAG setups are usually additive, with documents going in and answers coming out, and removing stale material rarely propagates to anything the system has already produced. Automatically rewriting or removing affected pages lets the knowledge base degrade gracefully instead of silently going rotten.

There are caveats worth noting. The version number, 0.9.0, signals a project that has not yet reached a stable 1.0 release, and the details here come from a single dev.to post rather than Tencent's own documentation, so anyone evaluating the tool should confirm specifics against the repository. Even so, the core design, one corpus with three query paths and automatic page maintenance, is a coherent response to a problem most organisations will recognise.