MediaWiki Chatbot Extensions Compared

Why Wiki Search Falls Short

MediaWiki's built-in search is keyword-based. Users need to know the right terms and where to look. When they don't, relevant content goes undiscovered, and the wiki's value as a knowledge base diminishes.

Several MediaWiki extensions now provide AI chatbot capabilities, each with a different architecture and trade-offs. We created one of them (AI Assistant) and have evaluated the others. Below is our comparison.

AI Assistant

AI Assistant is a chatbot widget that sits on every page of your wiki. Users ask questions in plain language and receive answers drawn from wiki content, with numbered source citations linking back to the pages used.

The extension is permission-aware: answers only draw from content the current user is allowed to see. It supports conversation history (save, browse, and resume past conversations) and works across languages, allowing users to ask in one language and get answers derived from content in another.

Configuration happens through the admin panel, where wiki administrators set the AI provider, customize the chatbot's appearance and system prompt, and control which parts of the wiki the AI searches. AI Assistant supports Anthropic Claude and OpenAI ChatGPT out of the box. Other providers, including locally running models, are available on request.

We developed AI Assistant as a commercial extension. It is available on ProWiki, our managed MediaWiki hosting platform, and can also be installed on self-hosted MediaWiki instances.

Wanda

Wanda is an open-source chatbot extension that uses Elasticsearch for content retrieval. It provides both a special page and a floating chat widget.

With Wanda, you get the broadest range of out-of-the-box supported AI providers: OpenAI, Anthropic Claude, Azure OpenAI, Google Gemini, and Ollama (for self-hosted models). This makes it a flexible option for organizations with specific provider requirements.

Setup requires an Elasticsearch server: CirrusSearch (the Elasticsearch-based search engine used by Wikipedia).

Currently, Wanda uses text-based Elasticsearch search rather than vector/semantic search, which means its content retrieval is less precise than vector-based approaches. Vector search is planned but depends on MediaWiki adding support for Elasticsearch 8.

KZChatbot

KZChatbot is an open-source extension built for Kol-Zchut, an Israeli rights information platform. It provides a floating chatbot widget with source citations and conversation history.

Unlike most of the extensions here, KZChatbot is a middleware layer. It handles the user interface, session management, and rate limiting, but relies on a separately deployed RAG (Retrieval-Augmented Generation) backend for the actual question answering. The reference backend is Webiks-Hebrew-RAGbot.

This architecture means KZChatbot is provider-agnostic (the backend determines which AI model is used), but it also means you need to deploy and maintain two separate applications. There is no built-in authentication between the extension and the backend; security must be handled at the network level.

KZChatbot includes sophisticated operational tooling: rollout controls, banned word filters, and batch testing with CSV export. It is licensed under GPL-2.0. While technically language-agnostic, it was built with Hebrew language support as a primary focus.

AskAI

AskAI takes a fundamentally different approach. Rather than automatically finding relevant content, it lets users manually select wiki pages and specific paragraphs as context for their questions.

The workflow is: search for relevant articles via MediaWiki's built-in search, click "Add to AI chat" on the results you want, optionally select specific paragraphs, then ask your question. This gives users full control over what the AI sees, but it also means relevant content the user didn't think to include will be missed.

AskAI only supports OpenAI out of the box (adding other providers requires writing PHP code), and it runs on a special page rather than as a floating widget.

AskAI is open-source under GPL-2.0 and is in early development.

Chatbot (OpenSemanticWorld)

Chatbot is part of the OpenSemanticWorld ecosystem for scientific knowledge management. The MediaWiki extension itself is an iframe wrapper that embeds a separately deployed chatbot application (osw-chatbot).

The osw-chatbot backend supports RAG and GraphRAG, making it potentially powerful for wikis with structured data. However, the MediaWiki extension itself does not handle content retrieval or indexing; that is left entirely to the separate backend. The extension is also tightly coupled to the OpenSemanticWorld stack and has limited documentation. Using it outside that ecosystem would require significant adaptation.

Chatbot is licensed under AGPL-3.0 and is in experimental status. In its current form, it is not a realistic option for most MediaWiki installations.

Feature Comparison

Setup Complexity

Setup complexity varies widely across these extensions and is often the deciding factor in practice.

AI Assistant requires minimal technical setup. On ProWiki, it is enabled and configured through the admin panel. For self-hosted installations, we handle the deployment.

AskAI is the simplest self-hosted option: install the extension and configure an API key. No external services required.

Wanda requires an Elasticsearch server, which is a significant infrastructure dependency if your wiki doesn't already use CirrusSearch.

KZChatbot and Chatbot (OpenSemanticWorld) both require deploying and maintaining a separate backend application in addition to the MediaWiki extension itself.

Open Source

All extensions except AI Assistant are open-source. Wanda, KZChatbot, and AskAI use GPL-2.0; Chatbot uses AGPL-3.0. The open-source extensions offer more flexibility and no licensing cost, but require more technical effort to set up and maintain. We offer AI Assistant as a managed, ready-to-use alternative.

Which One Should You Choose?

If you need a simple way for users to query specific wiki pages and don't need automated retrieval, AskAI is lightweight and easy to install.

If you have existing RAG infrastructure and need operational tooling like rollout controls and batch testing, KZChatbot's middleware architecture may fit your needs.

If you need self-hosted open-source and already run Elasticsearch/CirrusSearch, Wanda is a mature option with the widest range of AI provider support.

If you want a ready-to-use solution and prefer not to manage infrastructure, AI Assistant provides permission awareness, source citations, conversation history, and cross-language support with no setup required.

The AI and chatbot landscape is evolving quickly. If you come across something in this comparison that is out of date, missing, or otherwise inaccurate, let us know.

Get Started with AI Assistant

Want to add AI-powered search to your wiki? Learn more about AI Assistant or get in touch to discuss your needs.