If it chugs that badly with one user and zero load, I'm gonna assume it'll tip over if I so much as look at it funny.

SaaS is not the website itself. It's paying for risk management. For example, why did you use AWS instead of the wayyyy cheaper server-under-a-desk? At your scale, I bet it would be more than adequate.

If _you_ don't want to self-host something core to your business, think about why. Congrats, you now understand why people pay for SaaS.

Did you try it for yourself? The reason it's a bit finicky is that a lot of the actions in the chat skip the AI altogether, such as changing the cart values or changing the shipping options. In many other agentic commerce chat interfaces, every single action just involves sending a message in the chat. And then the AI calls the MCP action. But we thought, why not let some actions use the MCP tool directly? But this adds latency because you're interacting with the store's GraphQL API directly.

It’s still working for me and snappy but I’m in the region it’s hosted it. It’s also hosted on Railway so not sure where you got AWS from.

You don't gotta justify yourself. I don't really care why it's slow nor why you made the choices you made. And that's the point I was making: _neither will anyone else_. "It works on my machine" doesn't cut it when people are trying to determine if your stuff is worth the monumental risk.

I saw us-east and assumed it was us-east-1 but it doesn't matter. The same criticism applies to any hosting situation. You _could_ have gone with a cheaper DIY solution but didn't.

I appreciate your feedback and feel like optimistic refresh is probably the easiest fix for the issues here.

I’m curious what you mean by a cheaper DIY solution?