A: Hey Nirmantix!

Thank you so much for the question. You are right that allowing people to put their openai compatible systems would be much easier, but I don't do it for 3 reasons.

1. Google requires a whitelist of domains that will be receiving calls to and from SheetXAI. This already takes the entire self-serve out of the equation because I can't predict all the domains of all possible connections, so this will become a problem for customer support as people's models won't be working until I do the whitelist and manually redeploy the product.

2. Many API routers like Megallm, straico, etc, although great can be very limiting. I will give you an example, we are currently working on allowing streaming, tool calls, etc on SheetXAI, to allow the AI to be even more agentic. But since Straico (and many others) don't have streaming OR tool calls, we are essentially building 2 versions of the app, one for people who use Straico and one for everyone else. The worst part, the users who are on the non-streaming part would most likely notice sooner or later that the performance of their version of the app is not as smooth as the one for those who are using openrouter or openai, etc. Once again, night mare for customer support.

3. It's super hard to deal with changes to models when there are already 100s of them on our app, adding more can be a nightmare. Another example, some models are being deprecated on openrouter, openai, and Straico this month, the crazy part is that it's not the same models being deprecated, each one of them is having their own set of models that they are parting ways with. It's a hassle to have to constantly visit 10+ LLM websites every day or every week to check if they got rid of a model or not, then having to redeploy and explain to everyone why that model is missing.

bonus point 4: As you can imagine, we've already had many people try to steal our system prompts, i'm even sure that at least 1 person would have already succeeded by now haha. With the ability to bring their own compatible endpoints, this will literally be the easiest thing to do. I could build an openai compatible endpoint that just logs requests in minutes. Same for ollama and the others. So for security reasons, I also walk very carefully around these types of requests to make sure that we can securize it before allowing people to use it.

That said, I already plan to add deepseek, groq and a few others, no problem at all. I just have 500 things on my task list, a chrome extension coming in a few weeks, etc, and we are not a huge team, hence why I haven't added them yet since there is stuff that is already working. Because adding them is not just setting them up but we have to test them extensively to make sure they actually can work with our prompts. So, just bare with me, I'm working as hard as I humanly can.

For any feature requests, I'm always available at [email protected], so don't hesitate to email me. I promise I read every email and even reply to every single one. You can also add it to roadmap.sheetxai.com

Thank you!