Hello,
I am confused as to why I cannot get some basic functionality out of the assistant. As you can see in the screenshot I have exactly one file attached - it as about 1,000 lines of code - yet the Assistant is complaining about too many attachments. All I was trying to test was the simple prompt you can see in the photo with the one file. Thoughts?
Thank you,
Scott
i’m trying attach a script that was only 200 lines and the Assistant says he can’t read “attach” files.
the chat have a limit of characters so you cannot copy/paste your full code.
seems this IA was not made to read the scripts files on the project like you can do using vscode+copilot.
Thanks for the response. I had done a little more testing this morning and find as you did that it could handle only very small amounts of code at a time. The feedback on a small file was good, but I cannot be feeding it tiny - in my estimation anyway - bites of code to examine or refactor. Again, on the plus side, it seems to have a very good understanding of Unity code. Guess we’ll see how it goes. They most definitely need to make their errors/warnings more clear however.
Cheers!
i tested script with 580 lines and the AI cand red it complete. Scripts with 700+ lines is not possible.
the problem i found is, if you start split your script, the AI will end losing the context..
i think they need at least, put an option to AI get acess on your full script, no matter the lines, like you can do with copilot using gpt/claude/gemini on vscode.
i’m trying do things but im not good with codes, for now im my opnion i still think using IA on copilot is the better choose, since it can read your entire project scripts.
hope they look on it.
on my project right now (i’m doing a TCG card game), copilot AI cant help me fix one broken thing in the game logic, I would like to test the potential of Unity’s AI on this, but he can’t read the full scripts 
Thanks for letting us know. Context window size is always a challenge with LLMs, and we are doing a bunch of things to try and manage what goes into the context to make sure the assistant doesn’t forget critical things, and has the space to understand enough code at once.
This is a very active area of research right now, and we’re working improving, so thank you again for timely feedback.
Could you also try it with the /code route? do you get similar results?
I see that the object isn’t attached, you can drag and drop the file in the prompt area. but normally it should have picked it up, I will bring that up within the team. thank you
