The budgets of one chat.
Model calls per chat turn.
Context budget of the chat, in tokens (system prompt included).
Longest tool result kept in the conversation, in bytes.
Upper bound on generated tokens per call.
Sampling temperature.
The budgets of one chat.