Short Circuit is a $30 one-time purchase. Seems to fit your description/requests pretty well. I’m using the trial, have not purchased because I’ve been pretty happy with Viticci’s Shortcut solution which is also excellent but does not offer the features you’re looking for.
This fits most of your requirements except for the last one, which the developer promised but has not delivered yet:
It works rather well, there are a couple of bugs and issues here and there of course, as with many of the quickly developed AI chat apps, but at least this one is not a subscription and not a rip-off. It has pre-defined prompts (‘personas’) and lets you define custom ones as well.
Yeah, it seems a bit expensive to me too which is why I’ve not purchased it. That said, expensive is relative. Some seem happy to pay the increased subscription cost for Fantastical, no way I’d do that. But yeah, expensive. But I would also add that it seems like a pretty nicely designed app and fit the description of the OP. The ChatAI Unlimited also suggested above seems like a much better deal just in terms of cost.
For the time being have you checked out S-GPT by Federico Viticci from MacStories? He created something to use it with shortcuts. Doesn’t have everything you’re looking for but it is a start. The link is all his stories on it and the latest version of it.
I paid for this. $2.99 seems pretty reasonable. It’s a bit plain compared to Short Circuit but it works very well and at that price an obvious choice.
A bit of a side note, I spent 30 minutes using it and it seemed that at least 50% of the response was completely fabricated. There’s another thread on the forums and plenty has been said elsewhere about how much Chat GBT hallucinates/fabricates responses. For a couple of my queries I requested links and of 20 links for two different chats only 2 were real. The rest were fake. So, yeah.
With some very sharp people on this forum seemingly surprised that chatbots (in their present LLM implementation) do not give factual responses, I’m afraid that non-techies are being completely taken in by the plausible answers they receive.
After my post earlier today I thought I’d try using the chat in the Bing app to compare the results as well as the experience of the app. I think that it’s using the more up-to-date 4.0 model? Results are way better and include source links by default. The app itself isn’t too bad but would be better if it allowed for more customization. They throw up a sort of news page with lots of thumbnail links… the design isn’t too bad but the sources, yuck. But just clicking into the chat button is easy enough. And it allows for copying the results or sharing. Until the better 4.0 model becomes available to everyone this seems better.
Another way to consider is to use the OpenAI API with its Python packages in Jupyter notebook. Juno has included the package in its recent update. It’s fairly easy to make a chat bot and there’re tutorials to do this in 10 min.
Although this is a bit more technical, it provides a lot more flexibility.