Entonces, tampoco podemos usar Google? o stackoverflow? o un IDE?, ya puestos, ni un ordenador? y un libro? Yo me retiro ya de esta profesión y me dedico a plantar tomates, la verdad.

Coño, todo un emprendedor, como lo van a meter en la cárcel...
Nos esperan cuatro años en los que la humanidad se va a ver en una encrucijada.

The exchanged messages, being asynchronous and spaced out in time, require an initial anchor that the LLM gives you, an identifier, and if necessary, you have to provide all the previous context so that the LLM can resume the conversation with the human. Or with another AI.

The database must serve to save the state of the messages exchanged with the human actors and the AI, since you can't overload the AI ​​to save the different messages that can be exchanged between potentially millions of users at once and the AI.

All of this to serve as a distributed database for a solution that allows an AI to speak via text or audio messaging with different human interlocutors.

automatic synchronization for horizontal autoscaling using citus, monitoring, optimization, automatic sharding, index creation, and a few other things that I'm forgetting.