Hi all,
So between errors from WeWeb AI I tried to generate a few components.
The issue that I found is that I keep asking for a fix and the AI says its done, but its not included.
i.e. I have a tree dowp-down selector. I asked to an arrow for the users to use to get the dropdown values.
MY question is 
What is the best way to continue to iterate this? Keep asking over and over or is there a way t extract this and put into claude/cursor/chatgpt to fix it?
Thanks.
hi,
I think there’s no solid answer for that. Because with Weweb AI, you don’t have to do much. It handles all the changes in all element files automatically. So, sometimes it’s reasonable to make several attempts to get it working.
On the other hand, if the number of iterations > 5, then I would recommend migrating to external LLMs or IDEs. But in this case you should be pretty careful, as you need to explain the element files thoroughly to the LLM. And after that always do version comparison to ensure make sure that AI didn’t erase (or mutate) part of the existing code.
In this terms Cursor may be more comfortable.
Just use other AI tools and import the code into WeWeb. The current AI doesn’t really make it worth it.
So how would you export this? Each file at a time or is there a way to export/link this to vs code/cursor ?
Unfortunately there is not I think. You might wanna use VSC / Cursor as a starting point, at least I do it that way. I honestly have never used the WeWeb’s editor.
do we have a tutorial on connecting weweb to vsc/cursor ?
There is some videos around, but the WeWeb team never delivered on a proper documentation, so it basically is a few people who wasted enough time on it to actually know how to develop them locally. You can check out Youtube, https://developer.weweb.io/, or hmu for a session, if you wanna learn on my wasted time and mistakes :d
There’s a little hint: every time you ask Weweb AI to generate anythin it also generates the AI.md file with a detailed description of component that was created.
Feed LLM with all the files you get from Weweb AI (including AI.md) and it almost always shortens the path to desired code.
Interesting. So I would copy paste each one of these.
After that, do you just paste back in the weweb editor?
yep. And by this you create a new version of the component. So, all the versions are saved and you can switch between them if something goes wrong.
I find this hard to believe … because … WeWeb should already be sending that to the LLM. If not … I REALLY would love to see what their system prompts are because that makes ZERO sense to me that that would not already be included in the LLM call.
I seen the old version. It’s nothing fancy. They just tell the AI their conventions. Funnily enough. The prompt for the AI could have very easily be also turned easily into the dev docs, but they didn’t bother.
There is basically multiple prompts like this, talking about how to do a ww-config.js and how to interact with the WeWeb specific methods in Vue.js, after that, it’s just Vue.js development, there is also some details regarding that, because of course, WeWeb doesn’t work properly with the new Vue stuff, so they tell it to not use it.
I think they are sending that to Weweb AI. I was talking about the case when you use external LLM outside Weweb.
I see - that makes sense then.
Happy 2 year community anniversary btw, thanks for all of your contributions!
Just to give everyone an update—I’ve been experimenting with WeWebAI. As expected, it still runs into the usual issues. In my case, I was trying to work on a dropdown component and made about 20–30 calls or iterations. Unfortunately, more than half of those attempts returned compilation errors.
To work around it, I extracted all the relevant files, added some context, and moved the process over to chatGPT and Claude. From version 20 through to about version 60, I made edits one by one—copy by copy—until I finally got the result I was after. There were still a few errors here and there, but the end result was functional.
So yes, tools like Claude or ChatGPT can help, but the key is to make sure you’re giving it all the relevant files and context. That made a huge difference.
