Artificial intelligence (AI) and Moodle

AI Provider

Fri, 22 May 2026 22:04:17 GMT

by Eduardo Kraus.

Hello, Marcel.

I agree with your analysis, and for me the main point is that the problem is not only whether or not to create native AI Providers, but rather placing features inside Moodle core that depend on rapid evolution, while the reality of most Moodle installations is exactly the opposite, where updates are slow, difficult, expensive, risky, and often never performed. Many institutions spend years without updating Moodle, sometimes because of legacy plugins, internal customizations, critical integrations, lack of staff, validation costs, or simply because the current environment is working and nobody wants to risk breaking production. When a feature that needs to evolve every few months is tied to core, it is practically born with an expiration date.

In the case of AI, this becomes even more evident because models change all the time, APIs change, request formats change, and new capabilities appear, such as proper separation between system and user messages, structured output, JSON Schema, multimodality, tools, context control, and many other things. If the provider depends on the Moodle version, most institutions simply will not keep up with that evolution. In a short time, we will have Moodle installations using old, limited, and incompatible integrations compared with what the AI market will already be using.

When I first analyzed the ChatGPT provider, it seemed to work in a very simplified way, basically with a single prompt field. I do not know whether this has changed in current versions, but that design is dangerous. Anyone who works with AI knows there is a difference between system instructions, developer instructions, and user-submitted content. A student's answer must be treated as untrusted data and never as an instruction.

Now imagine an essay-grading plugin that expects to receive a JSON response with a grade and feedback. If the student's essay is mixed into the same prompt as the evaluator's instructions, the student can insert an instruction inside the essay asking the model to ignore the rubric and return the maximum grade. This does not even need to be obvious. It can be in HTML, in another language, in a fake quotation, in a footnote, in a hidden comment, or even in an artificial language. For Moodle, that is just part of the essay. For ChatGPT, it may become an instruction.

Updates are not a problem exclusive to AI. H5P is a great example. It used to be maintained as an external plugin, through mod_hvp, and could be updated separately. Now it has become part of core, with mod_h5pactivity and the Content Bank. This brought integration, but it also created a direct dependency on the Moodle version. Today, when a new H5P library requires a more recent API version, many Moodle installations simply cannot run that content. As a result, content producers need to think about the lowest common denominator, not the best available feature. Today it is at h5plib_v128, but the vast majority of Moodle installations are on older versions, which is why many content factories still prefer SCORM, because it will work.

Another point is that Moodle core should be leaner. Very specific features should remain as optional plugins that can be installed and updated separately. This does not apply only to AI Providers. It also applies to things like Safe Exam Browser, which I remember you mentioning. It is a very specific, complex feature, difficult to configure, difficult to support, and often ends up getting in the way of the student who simply wants to study and take the exam properly. It also does not provide the level of security many people imagine. If an institution needs it, fine, install it as a plugin. But it does not make sense for everyone to carry it in core.

The same applies to BigBlueButton. It can be useful in some contexts, but it should not be treated as something essential to Moodle. In many environments, people configure it, test it, use it for a while, receive complaints from teachers and students, and eventually move to Google Meet, Zoom, or another solution that is already part of the institution's routine. The feature stays there, but it does not solve a real problem for most users.

In my view, Moodle should have a more stable, minimal, and strong core. Core should focus very well on what is truly its purpose: courses, users, enrolments, permissions, grades, completion, APIs, and so on. Features tied to external vendors, specific tools, or fast-changing markets should be independent plugins.

That is why I agree with the main idea is plugins should be independent plugins, updateable through the marketplace, with their own evolution and with a much more careful security architecture. Otherwise, Moodle risks announcing AI as a novelty while delivering integrations that will become outdated before they are even truly adopted by most institutions.

Best regards,
Eduardo Kraus

AI Provider

Fri, 22 May 2026 12:13:32 GMT

by Marcel Almeida.

Originally posted at https://moodle.org/mod/forum/discuss.php?d=474227

I am a PHP developer and I teach PHP with AI at universities. At two of them, I developed a project to challenge the mandatory use of the Secure Browser imposed by the government, demonstrating its weaknesses and how, in practice, it contributes very little to the actual security of assessments. We are now in the final stage of getting it removed, since it is a plugin tied directly to the Moodle core and cannot be independently downloaded, updated, or removed.

At the end of the project, I was asked to analyze other aspects related to the educational ecosystem, and one issue especially caught my attention: the amount of time and effort Moodle is investing in creating native Artificial Intelligence providers. In my view, this represents a major waste of resources because Moodle is extremely difficult to update. I still have production environments running version 3.7 with no upgrade forecast whatsoever, and the most recent environment I use is version 4.1, which also has no update plans. IT teams face countless issues during the upgrade process, and in many cases the recommendation ends up being to create a completely new Moodle installation and abandon the old one because updating is too difficult, problematic, and often breaks things.

Meanwhile, AI evolves practically every day, and native AI providers integrated into the Moodle core depend directly on platform updates in order to receive improvements, fixes, and compatibility with new models, in practice, this means that most institutions will be stuck with outdated integrations just a few months after they begin using these features. The numbers themselves reinforce this view: according to stats.moodle.org, only 22.16% of installations are running version 5.0 or higher. In other words, out of every 10 Moodle environments, about 8 are significantly outdated.

Given this scenario, it makes far more sense for AI providers to be distributed independently through the Moodle plugin marketplace, that way, each institution could install, update, or replace providers whenever they want, without depending on a full platform upgrade.

Simply adding a few AI providers does not transform Moodle into a truly AI-oriented platform. Today, there seems to be an attempt to create the impression that Moodle has become AI, when in reality what exists are merely connectors to a few external services. That is still far from making the environment truly intelligent or integrated with the new reality of generative AI.

Being AI Friendly does not simply mean offering five or six configurable providers. it means rethinking architecture, pedagogical workflows, continuous updates, interoperability, personalization, and above all, the speed of adaptation to changes in the AI market - which directly conflicts with Moodle’s current update challenges.

AI-Provider at course level

Fri, 15 May 2026 10:39:43 GMT

by Rainer Schaufelberger.

Hi,

I am currently working with LeChat from Mistral, which I find very interesting. You can create libraries (including indexing websites), which, as I understand, allows you to build a RAG System. I have installed the Moodle plugin for Mistral AI as a provider and now have my own provider with all my libraries.

However, AI providers are not currently assignable at the course level. Given how AI is evolving, I believe teachers should be able to provide their own AI-providers (and libraries) within their individual courses.

I wonder, if this could be realized? Does anybody now?

Something Went wrong try again later Moodle AI 5.0.2

Wed, 13 May 2026 12:36:04 GMT

by Vahe Ghorghorian.

I am using OpenAI. created the provider and the placement.

added the key as well.

Upon using the generate text i get the following error

Something Went wrong try again later

Help Needed .the curl is giving the correct results.and models.

“Something went wrong” Error When Using AI Feature (Moodle 5.0.1+)

Wed, 13 May 2026 12:31:25 GMT

by Vahe Ghorghorian.

PLease anyone found a solution to this i am using the AI feature but i keep gettng something went wrong. i am using openai.

thank you

AI-Assisted Timetable Scheduling for Moodle — Looking for Community Feedback

Thu, 07 May 2026 14:18:13 GMT

by sameh naim.

what I found is a platform that does the scheduling https://timetable.digital
that has an api interface and I built a plugin that connect to it they offer a free version and it read moodle list of courses an cohorts

this is my plugin

https://moodle.org/plugins/local_timetable

AI-Assisted Timetable Scheduling for Moodle — Looking for Community Feedback

Thu, 07 May 2026 13:19:08 GMT

by AL Rachels.

Closest that I have come to anything similar is scheduled writing prompts for my Diary plugin activity. Setting the start and stop dates of each prompt is time consuming when starting a new academic school year. So, I guess you could say your item 4, would be the one I am most interested in.

Hmm, I am finishing up testing prior to release of Diary v5.1.0 which has a bunch of new prompt modes and completion metrics which just adds to the problem of getting in good valid dates. Makes me wonder, do you have a plugin for this yet?

AI-Assisted Timetable Scheduling for Moodle — Looking for Community Feedback

Thu, 07 May 2026 08:31:32 GMT

by sameh naim.

Hi everyone,

I’ve been working on an AI-assisted timetable and academic scheduling system integrated with Moodle, and I’d love to hear feedback from the Moodle community.

One challenge I keep seeing in schools and training institutions is that timetable management is still heavily manual:

teacher conflicts
room overlaps
workload imbalance
last-minute schedule changes
disconnected systems outside the LMS

In many cases, administrators are still relying on spreadsheets or standalone scheduling software with no direct Moodle integration.

The approach I’m exploring is:

AI-assisted timetable generation
automatic conflict detection
teacher workload balancing
room/resource optimization
synchronization with Moodle courses and users
support for Arabic and RTL institutions

The goal is not just “calendar management,” but reducing operational complexity for schools and making scheduling workflows smarter and more connected to the LMS environment.

I’m curious about the community’s experience here:

How are institutions currently handling timetable generation and conflict management with Moodle?
Do most schools prefer timetable functionality inside Moodle, or external systems with synchronization?
Which integrations would matter most in practice?
- attendance
- course access
- notifications
- mobile access
- teacher availability
- student groups
Has anyone experimented with AI-assisted scheduling workflows in Moodle environments?

I’d especially appreciate feedback from institutions managing:

large student numbers
multi-campus scheduling
teacher workload distribution
higher education scheduling constraints

Looking forward to hearing your thoughts and learning from existing approaches in the Moodle ecosystem.

Moodle's built-in analytics is good, but is it enough at scale?

Thu, 30 Apr 2026 09:29:48 GMT

by Rushikesh Devre.

Hi all,

I know Moodle has learner analytics and configurable reports, and for smaller sites they work well. This question is more for people managing larger installs, say 2,000+ users.

The pattern I keep hearing is this that the data exists, the reports are there, but someone still has to decide which 20 students out of 3,000 need attention this week. And that triage layer is where things fall through the cracks. Not because the data isn't accessible, but because at scale there's simply too much of it to review proactively.

One admin described it as: "I'm not missing data. I'm missing a way to know where to look first."
So I got curious for those of you running larger Moodle sites, how are you handling that prioritisation layer? Are you relying on the built-in alerts, custom reports, a third party plugin, or mostly reactive at this point?

Genuinely trying to understand how people are solving this in practice.

Cursive authorship and writing plugin

Thu, 23 Apr 2026 12:25:23 GMT

by Joseph Thibault.

I'm pleased to release Cursive 2.1.5:
Major Features
Student Dashboard - one of the best things about writing data is that it helps to quantify the toil, difficulty, and time that writing takes. We think students and teachers should celebrate writing just as an athlete might celebrate their personal best. Our new student dashboard brings an updated look and feel to writing history on Moodle, including metrics over time and assignment, aggregate stats, and more.

Student-facing Analytics - transparency is foundational to a culture of integrity. Teachers now have an option to provide student access to analytics for each assignment, rather than limiting them to the student’s profile (above). When enabled, students can view their own analytics from discussion boards, on the lesson review, assignment page, or quiz review.

Even Better Autosave [free] - During an internet disconnection, we actively ensure that any information from the student writing process is buffered until the internet is restored (as long as it’s restored while the student’s tab is active. This means less lost data, more automatic saves, and complete Cursive analytics and replay for every submission, even in high-latency or laggy connections.

Completely Remove Assignments [free] - previously, Cursive data was tied to the assignment module even if the student’s submission was deleted by a teacher for a re-attempt. As requested, we’ve tied Cursive analytics and replay information to the student submission instead, so a removed attempt will remove Cursive analytics and replay now giving the student a fresh submission.

Minor updates
Paste Count [free]: Even if pastes are allowed, display the number of pastes on the “difference” page (this is similar to how we already treated pastes within the Replay feature).
Writing Report Improvement [free]: Quizzes with multiple essay type questions now are now titled with their question titles in the Writing report for easy filtering and identification.
Cite-Source Comments in Replay [free]: Student comments from cite-source (an activity-level setting) are now included and assigned to individual pastes on the Replay page.
Bugs, stability, optimization - under the hood, we continue to push the envelope for performance and resilience