Opencode reading file again and again and fill context.

So I am using 3.6 35B A3B, pretty good for my work, but the first 64k tokens feels bigger like not a problem, but second time onwards it starts to read every file again and fills up context then context is emptied then it tries to read files again and so on, so no production after that, so what is solution of this, or do I have to start new session every time if so then how it gonna know about project and it will still feel the context so pls mention possible solutions.