Gemini (Google) has a good-enough-for-personal-projects free tier. Openrouter has free models (I'm assuming the training data is valuable enough to keep this going for a while)
There's a company (http://turso.tech) that can host sqlite as a service. One of the features allows a replica local sqlite file to be used for reads and the writes still go to the primary on their edge network.
Been using it (without the local feature) for a couple of months on a project and I will do it again for future products without a second thought.
consider not storing running_balance (or at least expect it to be written to a LOT). Based on the above description that's going to be a hot table with a write/update of 1 per second per row (of active customers).
If you are thinking sql, that running balance could be calculated from a VERY tall and skinny table of time_spent per session (which would be calculated in the application) or an append only table of (billing_account_id, timestamp) where you write a new row every second.
I use mermaid and flowchart in github readmes using an extension for chrome. The basic idea is that the libraries are part of the extension, and the markdown calls out what kind of diagram it is.
I used Dokku (and Dokku-alt) on Digital Ocean with the goal of making deploying easier. I ended up getting rid of it (and Docker) because for a small scale system, Docker wasn't worth the overhead (vs capistrano and ruby, or heroku's git push). On the other hand, at my last job -- where scale was much larger, we implemented a Docker orchestration system to handle our SOA using Consul as a registration system.
I saw a presentation on this last month, and found myself wanting to invent a problem just so that I could solve it with this type of scaling. Great presentation and great product.
make sure to document well. When you're scaling up a team and moving from a coupled code base to independent codebases a good amount of assumption happens, especially around how mature each codebase is.