I want to use
@Maple AI for coding, but it still has some maddening limitations that make it not a viable option yet. I don't know which are the LLM's fault and which are Maple's, but a few improvements would make a big difference:
File sharing:
- I can only upload one file at a time, and not with drag and drop - there's a file picker that starts from scratch every time instead of opening the last directory I used.
- There's no option to show it a repository online or a folder on the device so it can scan more than one file at once. Even without this, uploading a bunch of files in one message and doing it more efficiently would be enough.
Behaviour:
- No matter what I tell it, the AI responds to every prompt by jumping straight into fixing or writing code. I literally just said (using Kimi K2 Thinking), "I can only share files one at a time, so don't respond until I've shared a bunch to show you what's happening." I expected some short response back, maybe with observations about that first file, but sure enough, it generated a whole page worth of code. How do I get it up to speed with what I want when that's the response to every prompt, and it takes 10 prompts to get my point across? And all the unwanted processing takes up usage credits and context window space?
- I saw the same behaviour with Qwen3 Coder a few weeks ago - if it's a Maple thing, maybe it can be easily fixed.
With Claude Sonnet 4.5, I can tell it to wait as I get it up to speed, upload 20 or so files at a time if I like, and let it observe as I go. It does respond to each message, but it will wait if I say to until I tell it what I'd like to build next. Feels like a functional conversation with a capable employee instead of a frustrating conversation with an overconfident, pushy one.
I hope this is constructive - the tool is so close to being ready, I hope it becomes the one I use soon. It's fine for asking questions, but for working on a project like I am, it feels like a battle the whole way through.
Read the full announcement: 
