Yes, I think both features are worth trying. Let me try to explain.
When you click on Explore GPTs in the sidepanel, there is a button on the top right that says + Create, which takes you to the GUI you see in the picture below.
I don’t think you need “Actions” for your use case, but you do need the Code Interpreter & Data Analysis function. Only select what you really need – this will make the system prompt shorter, and I find it makes the custom GPT more focused. Sort of like “less is more” but only if you know what you want.
You can try uploading the HTML files to the knowledge base, but you could also create system instructions so it knows what you need right away, and upload them individually. If you always need the same thing, you could even add a Conversation Starter, which is just a button to press instead of typing the same prompt – just specify what these commands do in the system Instructions.
Custom GPTs don’t access any of your memories that were stored by vanilla ChatGPT, and they don’t access personalisation instructions.
And Projects are a bit different. Instead of having a specialised GPT that you can use over and over again, you would use a vanilla GPT (usually 4o) for a larger quest.
The custom instructions (personalisation) you can set in your account settings are also skipped, but the models can access your memories (and store new ones) and you can give them a custom system instruction per project, as shown in the image below.
In Add Files you upload your data for this project, and in Add Instructions a system prompt that will be loaded in every new chat within this project, and maybe… this way you can split the HTML source codes into several chats and still keep things OK-workable.
None of this is perfect. I wish we had much larger context sizes, but Sam Altman said in an interview that surprisingly this doesn’t seem to be a popular request – Personally, I’d love to have that more than anything. 4o is my favourite model.