How are system prompt treated actually?

In assistant API There are several input prompt
For example user, content,system
But how they are treated actually
How they are feeded by the model??