Project title: Agentic editing capability for Wanda extension
Brief summary: Wanda is a MediaWiki extension that provides an AI-powered chatbot interface for the wiki, available at both a dedicated special page and in a floating chat widget accessible from every page. It can answer questions about the wiki content using various LLM providers and includes. Wanda also provides an API that allows other extensions and services to interact with the outside LLM as well, serving as a bridge based on the wiki content. This project will add to Wanda the ability to edit wiki pages for a variety of different purposes.
Expected outcomes: Wanda should execute tasks on the user's behalf with appropriate permission checks in place - For example, "Do a Google search to find five Thai restaurants in Sao Paolo, Brazil, and make a bulleted list of their names, with each name as a link, in a new page called 'Thai restaurants in Sao Paolo' "
Skills required/preferred: PHP, Vue.JS, prompt engineering, LLM context management, basic knowledge on MW core API actions
Possible mentors: @Techwizzie , @Yaron_Koren
Expected size of the project: 350 hours
Rating: Hard
Microtasks: T407186, T413383
Any other additional information for contributors: T409213 has more details on the various scenarios expected.
Code - https://gerrit.wikimedia.org/r/q/project:mediawiki/extensions/Wanda.
Documentation - https://www.mediawiki.org/wiki/Extension:Wanda
Why are you proposing this project? Editing capability is one of the last major gaps in Wanda's functionality. Enabling it would dramatically improve Wanda's capabilities, allowing it be used as an editing assistant, helping it to automate major tasks like proofreading, research and restructuring of content.
What is the expected impact? Potentially, this project could have a very large impact, in providing the first-ever built-in AI-powered editing agent for MediaWiki. It could greatly increase the productivity of MediaWiki editors and maintainers. It could also serve as aa proof-of-concept for Wikipedia and other Wikimedia sites to provide their own agentic editing functionality in the future, whether or not they directly used Wanda.
Description
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Open | LGoto | T414121 Projects and mentors for Google Summer of Code 2026 | |||
| Declined | None | T414281 GSoC 2026: Agentic editing capability for Wanda |
Event Timeline
Hi @Techwizzie @Yaron_Koren thank you for this proposal. However, our ML team has assessed it and recommended that we decline it. This work is very complex and too large for the scope of the GSoC program. Such a large goal would not be setting a contributor up for success, so I'm afraid we cannot accept it.
We encourage you to consider revisiting this idea with a smaller project or related work that would be more suitable for a limited time frame and a less experienced developer.
@LGoto - I don't know who those ML experts are that you consulted with, but I believe adding agentic editing to Wanda is completely feasible. There are already agentic editing tools for MediaWiki - here is one example. (You can see a demo of it here.) For that matter, there was already a patch to add LLM-based editing to Wanda, here - it was abandoned for various reasons, but I tried it while it was still under development and it could in fact do basic editing, like "Simplify the wording in this page". What specifically did they think was unfeasible? If they want, we can reduce the stated scope - it's not a problem; I don't think anyone expects this single project to result in a complete agentic editing solution.
Hi @Yaron_Koren We do not believe that adding agentic editing to Wanda is unfeasible. Rather, our concern is that doing so in a way that would be both useful and aligned with community values would be a significant project that would require coordination with a range of teams within WMF, including design research, ML, research, product teams, and legal. Given this, we consider the project too complex and too large to be an effective GSoC project.
@LGoto - thanks for the response; but I don't understand this. Why would any of this project require coordination with the WMF?
@LGoto This project does not involve any fine-tuning or LoRA adaptation techniques. It consumes an enterprise / Ollama based LLM API to perform all the mentioned operations. The maximum knowledge that is expected of the GSoC student will be in prompt engineering (which everyone knows to an extent), classic MediaWiki extension programming (Vue, PHP) and core MW API actions.
In fact, this will be one of the best projects for someone entering MW development as a newcomer since the biggest success in this project relies majorly in writing good prompts and understanding how MediaWiki pages are managed.
Hi @Yaron_Koren and @Techwizzie as the sponsoring organization for the program, we are responsible for selecting projects that will provide student participants with a useful learning experience. Given the complexity and likely community controversy around this kind of project, we do not believe this proposal would enable the participant to succeed. Thank you for understanding.
@LGoto - sorry, I still don't understand. What community controversy? Is there some sort of general anti-AI sentiment even for non-Wikimedia uses of MediaWiki? And if so, does that mean AI-based editing as a general topic is off-limits for all future mentorship projects?
@Yaron_Koren The main problem for communities is WMF supporting an agentic editing project in any form. If you were to pursue this project on your own, people might disagree with you; but if WMF is involved (which we would be implicitly just by facilitating this proposed project), then the student participant would be facing a much bigger controversy. (If you're not familiar with these complexities, please see https://www.wikidata.org/wiki/Q130713537#sitelinks-wikipedia for a glimpse of the years of problems that LLMs have already created for the Wikimedia communities.) While we would not be involved directly in that community conversation, we cannot risk it for a new participant who is here to learn. Above all, our priority is ensuring a productive and safe learning environment for the participant.
I’m sorry that we cannot provide you with long discussions but we are a very small team and are actively managing multiple other projects.
Thank you; I wish you had just said this at the beginning, instead of all the stuff about programming difficulty and the need for coordination. I still disagree, though. Wanda is an AI tool that's restricted to the wiki that it's running on; it can't be used to edit Wikipedia. That's in contrast with other AI tools like MediaWiki MCP Server (the one I linked to above), which I believe can already be used to edit Wikipedia. So there's no reason why anyone would object to this particular project, and certainly there's no reason why the student doing it (who of course didn't conceive of the project) would get any criticism. On the contrary, I wouldn't be surprised if there are 10,000s of budding programmers out there who would jump at the chance to get hands-on experience with agentic editing, controversy or no.
I really hope you reconsider, because we are actually doing agentic editing the "right" way, and this project could end up being a big win, for both the MediaWiki community and the student in question.