Hey everyone! We have a significant update to share: we’re making some changes to our docs AI, and while it’s being sunsetted, we have some great updates ahead!
After evaluating its costs and the value it provides, we’ve found that many questions can be addressed more effectively through forums or by using a powerful LLM like ChatGPT or Claude. For those who enjoy using an LLM as a question bot, we’ll soon be providing a helpful guide on how to integrate and use our docs with an LLM easily.
Additionally, for anyone who hasn’t explored our docs search much recently, we have great news: @italojs has done a fantastic job revamping it! If you’ve encountered any issues with the search experience, there’s an open issue where you can report them. Feel free to create a new one if needed, and we’ll provide the support you need.
We’re confident these changes will lead to an even better, more efficient experience for everyone!
I have found the M3 Docs AI to be very useful – but if the same info is available via ChatGPT or Claude, that is just as good. Have they already scraped Meteor docs and forums?
no, not yet, but I plan on making a page with all the content of the forum so you could give the page’s URL to a LLM researcher
A prompt like this could help:
# URL-Based Retrieval and Question Answering Prompt
## Objective
You will be given a URL that contains a webpage or article. Your task is to:
1. Open the URL and extract the most relevant information that answers specific questions.
2. Summarize the key points from the content on the URL, focusing on accuracy and relevance.
3. Respond to user questions based on the information you find on the page.
4. Ensure that your answers are grounded in the content of the URL and provide citations or direct references to sections of the webpage when appropriate.
- **URL:** `<URL>`
- **Question:** `<question>`
### Response Instructions:
- Retrieve the page from the provided URL.
- Find and summarize the key features or relevant details described on the page.
- Answer the question based on the information provided.
When the AI search in the docs came out I briefly tested it but never used it again. The Docs are pretty well structured (at least for a human that browses the page) and if you have been working with Meteor for years you pretty much know where to look anyway. Different story for Coding Assistants though. I am using https://www.continue.dev with VS Code and when you give it n URL you can have it scrape the docs it finds there and make it accessible to the assistant. The problem with the Meteor Docs is that it get’s loads of docs for previous versions. I just asked it what the Roadmap says and (even though I just reindexed it) it still thinks we are on Meteor 3.1.2
If the docs were set up in a way that you can provide an URL for one specific Meteor Version and then have the scraper just get those pages that would be huge improvement.
The docs AI was absolutely phenomenal. I was heavily using it. There are issues with the docs, for example broken links from outside of the docs linking to the old docs that no longer work but which the AI could easily help navigate you to. Another really great thing is that the AI would aggregate information from across the new docs, and the old Guide, which is super helpful.
Overall, I feel that this is a big usability loss, especially for people who know how to prompt AIs very well, and especially for people who don’t know what to search for to begin with.
The old Guide is not linked anywhere on the new website by the way! It is not listed anywhere here:
and there is no link anywhere here:
If you navigate to Meteor 2 docs, then you’ll find the Guide, but this is not obvious at all, especially for new Meteor users.
The “Meteor 2” Guide is still very much relevant to Meteor 3, and should definitely not be hidden under “Meteor 2”.
The AI was able to show me results from both the new docs and the old docs, which is incredibly helpful considering that the “Meteor 2” Guide has information still relevant to Meteor 3.
The AI is also really good at answering questions that are not necessarily obvious in the docs. For example, it was thanks to the AI that I was able to gind an example on how to manually clean up Blaze instances using plain JS, and therefore make this PR to make the docs more helpful:
The purpose of such a change as that PR is that AIs are very good at gathering information that is otherwise not easy to find.
I really feel that we should restore the AI, and also get it connected to the Blaze community site because the Blaze setup is still a default Meteor setup so it is a loss for people exploring Meteor and not being able to easily discover how to use the default setup(s).
Docs AIs are the future of documentation sites because they can be trained to answer questions on a specific topic better than a generic AI, trained to follow specific rules, for example to prioritize results on specific sites and pages.
Furthermore, 3rd-party AIs don’t necessarily have the best UX. For example, here’s a Claude.ai result that shows no links, which makes it difficult to navigate back into Meteor docs to see relevant parts:
There’s a lot more we can do with AIs. For example, Meteor’s docs site could be updated to have live code examples that can be forked into user accounts, with code snippets that can be generated and inserted by AIs to help people make examples, working demos, and even full apps. This could enable an era of code sharing among the Meteor community, which could lead to developing and hosting apps directly on Meteor’s platform (like Replit) with minimal effort.
What could help with this is to add features like an official UI library,
as it would mean that people would be able to fork UIs that are automatically hooked into authentication and share these UIs with each other, while using AI to help them make more faster.
Imagine, for example, UI pieces that do useful things are shareable posts on Meteor’s platform (like instagram/bluesky but for Meteor code), backed by the Meteor stack and with Meteor’s user accounts that is already in place.
With standard backend-connected UI components, people could start easily making all the bits and pieces
Have they already scraped Meteor docs and forums?