Cost Management¶

AgentProbe tracks token usage and costs across all agent executions, helping you manage expenses and set budgets.

Cost Calculator¶

The CostCalculator uses YAML-based pricing data to compute costs from traces:

from agentprobe.cost.calculator import CostCalculator

calculator = CostCalculator()

# Calculate cost for a trace
summary = calculator.calculate_trace_cost(trace)

print(f"Total cost: ${summary.total_cost_usd:.4f}")
print(f"Input tokens: {summary.total_input_tokens}")
print(f"Output tokens: {summary.total_output_tokens}")

The CostSummary includes a breakdown by model:

for model, cost in summary.breakdown_by_model.items():
    print(f"  {model}: ${cost:.4f}")

Pricing Data¶

AgentProbe includes built-in pricing data for common models. You can provide custom pricing via YAML files:

# pricing/anthropic.yaml
models:
  - model: claude-sonnet-4-5-20250929
    input_cost_per_1k: 0.003
    output_cost_per_1k: 0.015

Point AgentProbe to your pricing directory:

cost:
  enabled: true
  pricing_dir: pricing/

Budget Enforcement¶

Set per-test and per-suite cost limits with BudgetEnforcer:

from agentprobe import BudgetEnforcer

enforcer = BudgetEnforcer(
    test_budget_usd=0.50,
    suite_budget_usd=5.00,
)

# Check a single test
check = enforcer.check_test(cost_summary)

# Check entire suite
check = enforcer.check_suite(all_cost_summaries)

CLI Commands¶

Cost Report¶

agentprobe cost report
agentprobe cost report -a my-agent
agentprobe cost report -f json

Budget Status¶

agentprobe cost budget
agentprobe cost budget --max-cost 0.50 --max-tokens 1000

Configuration¶

cost:
  enabled: true
  budget_limit_usd: 10.00
  pricing_dir: pricing/

budget:
  test_budget_usd: 0.50
  suite_budget_usd: 5.00

Best Practices¶

Set budget limits to prevent runaway costs during development
Track costs per model to optimize model selection
Monitor cost trends using the metrics system
Use pricing overrides for custom or self-hosted models
Include cost checks in CI to catch expensive test changes