Skip to main content

Cost Tracking

Multi-level cost aggregation: per-token → per-task → per-session → per-project → fleet-wide.

Architecture​

Token Usage → CostEntry → CostEngine → Aggregation
                                ↓
                     session / project / fleet totals

CostEntry​

public struct CostEntry: Codable, Identifiable {
    let provider: AgentProvider
    let model: ModelIdentifier
    let inputTokens: Int
    let outputTokens: Int
    let cost: Decimal
    let timestamp: Date
    let sessionId: String
    let taskId: String?
}

CostEngine​

let engine = CostEngine(persistence: persistence)

// Record costs as they come in
engine.record(entry)

// Query at any level
engine.totalCost(forSession: "session-1")     // $0.45
engine.totalCost(forProject: "my-app")        // $12.30
engine.fleetTotalCost()                        // $47.82

// Analytics
engine.costPerProvider()    // [.claude: $30, .codex: $17]
engine.burnRate(windowMinutes: 60)  // $/hour rate
engine.averageCostPerTask()         // Average task cost

Pricing Model​

The TokenPricing struct implements per-model pricing:

ModelInput (per 1M tokens)Output (per 1M tokens)
Claude Opus$15.00$75.00
Claude Sonnet$3.00$15.00
Claude Haiku$0.25$1.25
GPT-4$30.00$60.00
Gemini Pro$7.00$21.00

Alert Thresholds​

public struct CostAlertConfig {
    let perSession: Decimal   // Default: $10
    let perDay: Decimal       // Default: $50
    let perProject: Decimal   // Default: $100
}