WeKnora: Tencent's Open-Source RAG & Wiki Platform Guide

WeKnora is an open-source LLM knowledge platform built by Tencent's WeChat team and released under the MIT License. It lets you work with uploaded documents in three modes: RAG search for question answering, a ReAct agent that autonomously investigates, and a self-maintaining wiki that grows on its own. The bottom line: it's a full-featured, self-hostable platform written in Go, best suited to companies that want advanced knowledge management while keeping data on their own servers or private cloud.

WeKnora originated as the internal technology powering Tencent's WeChat Dialog Open Platform, which lets businesses embed chatbots into WeChat Official Accounts. As of this writing (September 22, 2026), it has roughly 28,700 GitHub stars and its latest stable release is v0.8.0. The GitHub repository was created in July 2025 and its first release shipped in September 2025; in September 2026 it hit GitHub Trending and drew renewed attention. Development has been active, with releases shipping roughly once or twice a month since then.

WeKnora itself is free, MIT-licensed open-source software. Any environment that can run Docker and Docker Compose — your own servers, a VPS, or a private cloud — can be up and running in minutes. What you do pay for is whichever LLM API you connect (OpenAI, Claude, DeepSeek, etc.) and the cost of running your own infrastructure. Using Ollama for local models can eliminate external API charges entirely.

Once started, visit http://localhost to access the web UI. To use a local Ollama model, run ollama serve > /dev/null 2>&1 & first. Extra components like Neo4j for the knowledge graph, MinIO for object storage, and Langfuse for tracing are each enabled independently via --profile flags.

After logging into the web UI, create a workspace and a knowledge base, then upload your internal documents. WeKnora supports 10+ formats — PDF, Word, Markdown, HTML, Excel, PowerPoint, images, and XMind among them — and uploading a whole folder preserves its directory structure.

WeKnora also ships as an MCP server, so Claude Desktop or any other MCP-compatible client can search and register knowledge in a WeKnora knowledge base (see the MCP configuration guide). For API integrations, scoped API keys let you restrict programmatic access to specific capabilities.

Beyond Docker Compose, WeKnora is also distributed as a Chrome extension, an npm package (@wxg-prc-cpg/dsh-weknora), and a ClawHub skill, giving you lightweight ways to call it from a browser or an existing chat tool without standing up the full stack yourself.

Because people often search for comparisons like "WeKnora vs Dify" or "WeKnora vs RAGFlow," here's how it stacks up against common alternatives. These are qualitative comparisons, and features on all of these tools change frequently.

- Login authentication has shipped since v0.1.3, but the maintainers themselves recommend running production deployments on an internal network rather than exposing them directly to the public internet
- The knowledge graph feature requires enabling the Neo4j profile separately — it isn't part of the default setup
- The project ships new features quickly, so the recommended upgrade practice is to pin WEKNORA_VERSION in .env before running docker compose pull
- A Japanese README (README_JA.md) is available, but the detailed product documentation is mostly in English and Chinese
- Self-hosting assumes both Docker knowledge and available server or cloud resources

- Good fit: companies that want to self-host internal document search, teams that want to mix and match multiple LLM providers and vector databases, and IT staff who want agent and MCP capabilities consolidated into one platform
- Not a good fit: individuals or small teams who'd rather avoid server operations and stick to a cloud SaaS, or organizations without the resources to configure Docker and LLM APIs themselves

A common pain point for IT staff at small and midsize companies is that internal regulations, manuals, and old quotes are scattered across a file server and hard to find when needed. Uploading these documents into a WeKnora knowledge base folder by folder keeps them searchable while preserving the original folder structure. Fixed-format documents like work rules or approval templates are a good match for RAG Q&A mode — asking "what's the carry-over rule for paid leave?" gets an instant answer — while messier material like old quotes, where conditions vary case by case, suits the ReAct agent mode for compound requests like "find past quotes with similar conditions to Company A's and compare them." Running a batch of manuals through wiki mode turns them into an interlinked knowledge base that keeps growing with every update. Reducing the "where was that document again?" problem is especially valuable for companies without a dedicated IT department.

WeKnora is an open-source platform that consolidates RAG, agent, and wiki-style knowledge use into a single self-hosted system, built on technology Tencent's WeChat team refined in production. Being MIT-licensed and self-hostable, with support for local LLMs and private vector databases, makes it appealing to companies that don't want data leaving their own infrastructure. At the same time, it assumes you're comfortable managing Docker operations and security settings yourself, so it's worth deciding upfront how much of the operational burden your organization wants to own. For more on RAG and knowledge-base design, see our articles on building internal knowledge search with Qwen3.5-9B and RAG and building a RAG-powered customer support AI with Ollama and OpenClaw. If reducing vector search costs is a priority, our complete guide to Amazon S3 Vectors is also worth a look. We also support companies with adoption and team-building through AI consulting and AI BPO — feel free to reach out via contact.