pub fn token_budget_base_overhead(
config: &TokenBudgetConfig,
system_prompt: Option<&str>,
) -> usizeExpand description
Token estimate of the fixed content every window starts with: system prompt + window info + guidance + seed (no thread hint — the best case; app-provided hints add on top).
Call ONCE at build time — the system prompt is process-constant, so the result never changes for the lifetime of the agent. The estimate is a char count (~4 chars/token Latin, ~1.5 CJK), not a tokenizer call.