For those that are interested (I might think of a few…somewhere), here is a link to one of the engineers that built Replit Agent:
Replit Agent - IDE for Humans and LLMs
It might give some relevant insights to current development projects and discussions about those projects.
All subscription in AI is about $20 a month. An idea of the reason?
Short answer, no. Probably some research on what a large number of people would be willing to pay. OpenAI seems to have started it and everyone else just used them as the example.
If you really want flexibility and you have the skill to use them, API’s are the way to go for a lot of functions.
However, while Apple is leading the way with AI on the devices, there are many companies pursuing that same goal. You can even get Raspberry PI’s with AI chips and AI-specific hardware. If Apple can shake the subscription temptation, I would love to see them release dedicated AI boxes for the home and office. I would love headless boxes (anywhere from Apple TV to Mac mini sizes) running Apple Intelligence (the one I hope we get and not this anemic, 2 years behind, poor Grammarly version) safe and secure in my home and office.
My biggest gripe is the limitation on the size of the files output, response. Are you a subscriber? Does a subscription allow for larger, longer responses?
Not sure to understand, for text it can give me forty lines I don’t need to correct, but maybe we don’t talk about the same subject 
When I use it for writing, it limits me to about 1500 words in an answer. A lot of my writing can be up to 100,000 words.
ha ok. So the answer is no, generating a few thousand characters is a limitation. You have to proceed sequentially to obtain longer texts. This limitation echoes the concerns of human writers (formerly called writers, authors, or even artists) and their rights over their work. More generally, the generative possibilities of AI are limited, on the one hand, to avoid too massive a shield raising, and on the other hand, to have time to discern how far this AI can take us. The current limitations that are often presented as technical are mainly linked to these elements. Nothing technically prevents gpt from writing a 300-page text since it is capable of providing me with the summary of a set of my texts (around 500 pages) in a few seconds. Moreover, automation removes this limitation, but this is an area that I do not master, @jscotta will undoubtedly be able to enlighten us.
@Flash what tool are you using (I do not want to assume ChatGPT). If you are not a subscriber, then you will be limited in your access to various models and their capabilities.
There are several factors that will impact your results on any of the AI services:
The subscription level
Your prompting skill
The overall service (e.g. Claude, Grammarly, ChatGPT, etc.)
Using the simple chat service, using an AI assistant, or custom service with API access for the AI power
Large amounts of data in or out has different results for each of the approaches. Also, the types of data. You say you write so it is factual or fictional? How was it prompted?
If you want to DM or email me, I can give more specific advice without wearing on Realmac’s open forum, too much (I already try their patience…and it is much appreciated).
Bruno is correct. Generally, breaking down a task into smaller chucks is better for working with an LLM–just like it is for humans. That is why books have chapters! Makes for better organization and helps us with processing. Same with GenAI.
That answers the question. Thanks wonderful people.