ExAgent.CostGuard (ExAgent v2.1.0)

View Source

Helpers for turning token usage into an estimated cost, used together with the max_budget_cents field of ExAgent.UsageLimits.

ExAgent does not ship a pricing table (model prices change constantly and vary by vendor). You bring the prices that matter to you as a map and CostGuard.estimator/1 builds the (usage -> cents) function you pass to ExAgent.run/3 via the :estimate_cost option.

Example

pricing = ExAgent.CostGuard.estimator(%{
  input_per_1k_cents: 250,
  output_per_1k_cents: 1000
})

ExAgent.run(agent, "go", estimate_cost: pricing)

Summary

Types

pricing()

Functions

estimate(estimator, model, usage)

Evaluate legacy homogeneous or model-aware pricing without losing unknown/error states.

estimator(pricing)

Build a cost estimator (Usage.t() -> fractional cents) from a pricing map.

Types

pricing()

@type pricing() :: %{
  optional(:input_per_1k_cents) => number(),
  optional(:output_per_1k_cents) => number()
}

Functions

estimate(estimator, model, usage)

Evaluate legacy homogeneous or model-aware pricing without losing unknown/error states.

estimator(pricing)

@spec estimator(pricing()) :: (ExAgent.Message.Usage.t() ->
                           non_neg_integer() | float() | :unknown)

Build a cost estimator (Usage.t() -> fractional cents) from a pricing map.

Prices are per 1000 tokens, in cents (250 means $2.50 per 1K, or $2500 per 1M tokens). Fractions are preserved rather than truncating every small request to zero. Missing rates remain unknown when their token direction is used; they do not imply a free provider. This is a flat input/output approximation, without cache discounts or adding reasoning subsets twice. Supply a model-aware callback for differentiated prices.