WeKnora converts your team’s documents into an AI-powered knowledge base that you can query and connect to your agents.
For an engineering team, a good starting point might be a collection of architecture decisions, deployment guides, onboarding documents, and incident reports. The questions are familiar: Why did we choose this approach? What should we check before publishing a release? Where’s the procedure that someone wrote six months ago?
WeKnora offers a shared interface for searching through that material and getting answers. Tencent’s open source project also includes an agent mode and a system that generates linked wiki pages from documents. Official Repository
What is WeKnora and what is it for?
WeKnora is a platform you can host on your own infrastructure to manage documents, query information through retrieval-augmented generation—RAG—and work with AI agents.
Its three main capabilities address different needs:
- Quick Questions and Answers: retrieves information from a knowledge base and uses it to answer.
- Agent Mode: addresses questions through multiple searches and available tools.
- Wiki Mode: transforms documents into linked Markdown pages that you can review and edit.
The project supports formats like PDF, Word, Markdown, and spreadsheets, plus integrations with sources like Notion and GitLab. As of September 18, 2026, GitHub identifies v0.8.0 as the latest published version. Documented Features · Published Releases
A useful first application would be to gather a single team’s operational documentation. Keep a scope that lets you verify whether the answers are correct.
How does an AI-powered knowledge base work?
WeKnora retrieves relevant fragments, incorporates them into the context of a language model, and generates an answer with references to sources.
Its documented workflow includes query processing, parallel retrieval, result reranking, fragment combination, and answer generation. Quality depends on what information reaches the model: a well-written answer doesn’t make up for retrieving the wrong procedure. RAG Pipeline Documentation
Think of a question like this:
What checks should we complete before rolling back the payment service deployment?
A useful result should point to the corresponding procedure and preserve its conditions and exceptions. During evaluation, review the cited fragment with the same care as the generated answer.
If you’ve already explored MarkItDown to convert documents into text that an AI can process, WeKnora addresses the next step: organizing that material, retrieving information, and querying it on an ongoing basis. You don’t need to add a separate conversion for each format that WeKnora already supports.
How to install WeKnora with Docker?
The official guide uses Git, Docker, and Docker Compose.
First, download the project and create its environment variables file:
git clone https://github.com/Tencent/WeKnora.git
cd WeKnora
cp .env.example .env
Before starting the services, review .env and replace the example credentials and required values. The file documents database options, authentication, storage, model connections, and optional integrations. Don’t commit it to version control. Environment Configuration
Next, pull the images and start the containers:
docker compose pull
docker compose up -d
The default address for the web interface is http://localhost. These commands follow the project’s guide and use the current repository configuration; they do not pin the deployment to the revised version for this note. Installation Instructions
The steps were verified against the official documentation, without running an installation for this article.
Having the containers running is the first step. You still need available models and an initialized knowledge base to query your documents.
How do you query your documents with WeKnora?
Create a knowledge base of documents, select a conversation model and an embeddings model, load some files, and wait for processing to complete.
The setup wizard lets you test the connection with the models before saving. The conversation model generates the responses; the embeddings model lets you represent the content for retrieval. The documentation warns that changing the latter after indexing requires rebuilding the index. You can add optional features later, such as result reranking or image understanding. Getting Started Guide
For a first evaluation, prepare three questions:
- A factual question whose answer appears explicitly in a document.
- A question that requires combining information from two documents.
- A question whose answer does not exist in the collection.
The third is especially useful. It’s worth observing how the system responds when it finds no evidence before the team starts relying on it.
This is a proposed evaluation method, not a claim that WeKnora has passed those tests with your documents.
Can you use WeKnora with Ollama?
Yes. WeKnora documents compatibility with Ollama and remote model providers.
Its system distinguishes between conversation models, embeddings, reranking, image understanding, and speech recognition. This way you can choose the services your collection needs, without assuming that a single model solves all tasks. Models Documentation
The important distinction is where each service runs. Hosting WeKnora in your infrastructure does not, by itself, guarantee that all fragments stay there: remote inference or embeddings services may receive content.
If you need local processing, review the complete journey, from file analysis through embeddings and response generation, before describing the deployment as local.
How do you connect your agents to WeKnora via MCP?
WeKnora offers an MCP server that exposes its knowledge management API to compatible clients.
The documented tools let you locate knowledge bases, perform hybrid searches, add documents, and converse about their content. This opens up a practical integration: an agent can retrieve team documentation while working on a task. MCP Server Documentation
For example, a development agent could consult a deployment guide before proposing changes to the publishing process. Whether that improves the outcome depends on the quality of the sources and how the agent uses the retrieved information.
The integration requires a running WeKnora instance and API credentials. Follow the project’s MCP setup guide, along with the current instructions for the client you choose.
Review which tools you enable: the server includes management operations in addition to search. Adjust access to match the task the agent should perform.
What does an AI-generated wiki provide?
Wiki mode produces persistent, linked pages from the knowledge base material, so you can browse it in addition to asking questions.
The project documents manual editing, version history, and version restoration. These are useful controls because generated documentation still needs a person responsible for reviewing terms, resolving contradictions, and correcting confusing explanations. Wiki Mode Documentation
For an onboarding collection, a reasonable test would be to check whether the pages clearly explain the relationships between services and whether someone new to the team can trace those explanations back to their sources.
Is WeKnora free and when should you try it?
WeKnora’s own code is licensed under the MIT License, with separate terms for included third-party components. Its operation still carries costs for infrastructure, inference, storage, and maintenance. License
It makes sense to evaluate it when a team needs to regularly query a collection of documents and wants to make it available to its agents as well.
Start with a knowledge base and questions whose answers you can verify. Measure whether people find the right procedure faster, whether citations support the answers, and whether maintaining the system requires less effort than the work it saves.
That is the test that turns an extensive list of features into a useful tool for software development.