Struct CostTracker 

pub struct CostTracker { /* private fields */ }

Available on crate feature eval only.

Tracks cost and latency metrics from agent event streams.

Uses per-model pricing tables to compute estimated USD costs from token counts extracted from Event streams.

Example

use adk_eval::cost_tracker::CostTracker;
use adk_eval::pricing::ModelPricing;

// Use default pricing
let tracker = CostTracker::new();

// Or provide custom pricing
let custom_pricing = vec![
    ModelPricing::new("my-model", 0.001, 0.002),
];
let tracker = CostTracker::with_pricing(custom_pricing);

Implementations

impl CostTracker

pub fn new() -> CostTracker

Creates a new CostTracker with default pricing for common models.

Default pricing includes Google Gemini, OpenAI GPT, and Anthropic Claude model families.

pub fn with_pricing(pricing: Vec<ModelPricing>) -> CostTracker

Creates a new CostTracker with the specified pricing table.

Arguments

• pricing - A list of ModelPricing entries to use for cost computation.

pub fn extract_metrics( &self, events: &[Event], duration: Duration, ) -> CostMetrics

Extract cost metrics from an event stream.

Iterates over events looking for UsageMetadata on LLM responses. Token counts are summed across all events that contain usage metadata. If no usage metadata is found, token counts default to zero.

The duration parameter is converted to milliseconds for the latency_ms field.

Note: The cost_usd field will be None because the model name is not available on the Event struct. Use compute_cost separately when the model name is known.

Arguments

• events - Slice of events from an agent execution.
• duration - Wall-clock duration of the execution.

Returns

A CostMetrics struct with aggregated token counts and latency.

pub fn compute_cost( &self, model: &str, prompt_tokens: u64, completion_tokens: u64, ) -> Option<f64>

Compute cost from token counts and model name.

Uses the formula:

(prompt_tokens / 1000.0) * input_cost_per_1k + (completion_tokens / 1000.0) * output_cost_per_1k

Returns None if the model is not found in the pricing table.

Arguments

• model - Model identifier to look up pricing for.
• prompt_tokens - Number of input tokens.
• completion_tokens - Number of output tokens.

Trait Implementations

impl Default for CostTracker

fn default() -> CostTracker

Returns the “default value” for a type.