So I tried to build a mobile app, and Bolt decided to use React Native.
I was having issues, some errors and the view was not working.
After trying to fix the issue someone told me that Bolt doesn’t support React Native.
But it was bolt that suggested using it.
What gives? This is a big WTF moment.
Some people can waste lots of time using Bolt for something that it doesn’t support but I feel like it’s not clear at all that it only works with some technologies?
To clarfiy => bolt.diy itself does not provide this answers. It´s just a wrapper that takes the informations coming fromt he chosen provider/LLM and writes in into files, as well as build it within the WebContainer.
So the output and quality depends highly on the chosen Model you choose and the prompts you give it build your app.
At the end it´s still needed you understanding how everything works together.
=> you wrote here that bolt suggestet it, what is defintively wrong. This comes from the LLM and not from bolt. So it depends on the user prompt and the system prompt, wheater it get to a result that is also working within the webcontainer.
I also dont know what exactly your expectation is with react native. the native is, as far as I know, what OS you got or want to build it for. So you could develop a webapp that can be previewed as webapp in bolt and also compile in productive to Android/iOS, but this native app you would not be able to run in the webcontainer.
Maybe its easier if you just share your chat history or your project, so we can rely on that and discuss on this specifically.
bolt.diy not limit anything, its just not optimized to handle/doing well with the responses for other languages (dont mixup languages with frameworks)
the webcontainer for the preview e.g. can just handle nodejs projects (state today. in the future maybe coming more like php, python, …)
It can be the case, that specific frameworks for JS also not working within the Webcontainer/Preview, but they should work when building/running it outside on the corresponding device (if you build react native android, you cannot expect it to work on windows running)
Bolt.diy can support any JavaScript based languages, frameworks, etc.
I don’t think that’s the issue you are having. I suspect that it’s more an issue with the model you chose. Even the best models make stupid mistakes, and a lot of those issues can be mitigated by better prompting. BE SPECIFIC with what the Agent should do, make your constraints clear, and what web-stacks/technology it should use.
And you want to make sure you use an AI Model that is good at code, such as Claude Sonnet 3.5 (what Bolt.new uses), DeepSeek-V3 (very good open source model), Google Gemini Flash 2.0 (good), or Qwen2.5-Coder-32B (minimum).
ALL AI hallucinate and skip steps, so you need to know enough to know when it’s lying to you, and you should have an understanding of the technology you are using. Otherwise, it will often cause you to make assumptions or troubleshoot the wrong issue.
P.S. Bolt.diy is open source, still very new, and Agent tools like this are all still a bit experimental… so using them is not for the faint of heart. I don’t even have enough patients for it at times; AI can be very frustrating.
Sadly, AI is honestly bad at programming, particularly at compound logic and OOP. And maybe that’s because of the training data… the average code at best is mid (often static and redundant).
Imo, AI needs a Knowlege Base (with smart memory retrieval through a Vector database), access to tools (browser, integration, etc.), automatic feedback (loop the response back to itself), a simple RAG (confirm criteria was met, expand the prompt/logic, determine criteria/etc., and break down the request into smaller steps as needed), and finally some sort of ground truthing. I’m personally working on a project to do right now but it’s proven to be a little complicated simply due to all the moving parts, but doable.