Description
JianShanAgent (见山) is an open-source QQ bot framework specifically developed for MediaWiki sites. It addresses the common challenge in community groups, such as game player communities or fan wikis, where users frequently ask questions about character stats, skill mechanics, event rewards, or seek information from screenshots. Traditional keyword-based bots are limited to fixed commands, while generic large language models often hallucinate or provide incorrect version-specific data.
Read More
JianShanAgent overcomes these limitations by integrating precise querying with robust large language model verification. It offers direct structured lookups using commands like `/查` (query) and `/筛选` (filter), which directly access MediaWiki APIs or structured site data. These commands return formatted cards within 1-2 seconds, bypassing the large language model to ensure data accuracy and prevent fabrication.
For natural language questions, the bot employs a multi-step verification process, combining Retrieval-Augmented Generation (RAG) with independent review. It first breaks down the question and identifies aliases, then searches local vector knowledge bases and real-time wiki content for evidence. After generating an answer, an independent review step filters out speculative information or unverified links, providing only grounded facts with source citations.
Key features also include reverse image search, which uses a vector model to index in-site illustrations, skins, and icons, allowing users to identify sources by uploading images. An hourly `kbsync` process automatically performs incremental updates to the knowledge base by checking recent MediaWiki changes. The framework comes with a pre-compiled web administration backend, eliminating the need for Node.js installation. This console allows users to view the complete retrieval and verification chain for each Q&A, debug conversations directly, and hot-update configurations and plugin switches.
Currently, JianShanAgent is deployed in production instances for Huiji Onmyoji Wiki and BWiki Onmyoji Story Encyclopedia, having indexed approximately 9,000 documents, 19,000 knowledge snippets, and 3,000 images. The project is entirely open-source under the MIT License, with its core framework, site plugins, and deployment templates available on GitHub.
- CategoryAI Chatbot Builders
- Pricing ModelOpen Source
- DeploymentPython Native, Docker Compose
- PlatformsMediaWiki
JianShanAgent's Core Features
Open-source QQ bot framework for MediaWiki sites
Precise command-based structured data lookups
Multi-step RAG with independent answer verification
Reverse image search for in-site illustrations and assets
Hourly incremental knowledge base synchronization
Built-in web administration console for debugging and configuration
Support for custom site plugins and extensions
Direct integration with MediaWiki API for data retrieval
Local vector knowledge base for efficient searching
Automatic alias recognition for natural language queries
Deployment options: Python native or Docker Compose
Template instances for quick bot setup
MIT License for all repository code
Supports multiple MediaWiki platforms (e.g., Huiji, BWiki)
How to use JianShanAgent?
Create Instance: Use the `template-instance` to create a new repository for your bot.
Configure Settings: Fill in your MediaWiki API address in `config/core.wiki.yaml` and large language model/QQ bot keys in `config/secrets.yaml`.
Choose Launch Method: Run natively with Python (`uv sync && uv run jianshanagent`) or use Docker Compose (`docker compose up -d --build`).
Access Web Admin: Open `http://127.0.0.1:8080` in your browser, retrieve the login token, and test Q&A in the 'Web Dialogue' section.
Build Knowledge Base: Execute `knowledge run / embed / load` commands in the admin console for full vector retrieval.
JianShanAgent's Use Cases
- Community Q&A Automation
- MediaWiki Data Retrieval
- Knowledge Base Verification
- Image Source Identification
- Wiki Content Synchronization
- Bot Management & Debugging
- Custom Wiki Integration






