Struct EmbeddingTokenBudget 

pub struct EmbeddingTokenBudget { /* private fields */ }

One execution’s embedding budget: tokens against its own ceiling and the server-wide one, plus the outbound request counter.

Implementations

impl EmbeddingTokenBudget

pub fn new(limits: EmbeddingLimits, aggregate: SharedTokenBudget) -> Self

pub fn unmetered() -> Self

A budget that never refuses, for runs with no operator-installed budget.

pub fn limits(&self) -> EmbeddingLimits

pub fn used(&self) -> u64

pub fn held(&self) -> u64

pub fn requests(&self) -> u64

Outbound requests counted so far.

pub fn reserve(&self, alias: &str, tokens: u64) -> Result<(), EmbeddingError>

Reserve a call’s whole estimate before dispatch. Like submilli:llm, refused once held reserve is already past its cap, so held spend per run is bounded by the cap plus one call’s sent estimate.

pub fn mark_sent( &self, alias: &str, estimate: u64, ) -> Result<(), EmbeddingError>

Count one outbound request and move estimate (already reserved) to held. Call immediately before sending; on Err, do not send. The error is a EmbeddingError::BudgetExceeded with no settlements — attach the provider’s with EmbeddingError::with_settlements.

pub fn settle(&self, reserved: u64, settlements: &[SubBatchSettlement])

Apply a call’s settlements and release whatever of reserved was never sent. Idempotence is not provided: call once per reservation.