orka.agents.local_cost_calculator module

Local LLM Cost Calculator

Calculates real operating costs for local LLM inference including: 1. Electricity consumption during inference 2. Hardware amortization (GPU/CPU depreciation) 3. Optional cloud compute costs

No more fantasy $0.00 costs - local models have real expenses.

class orka.agents.local_cost_calculator.CostPolicy(value)

Bases: Enum

Cost calculation policies for local LLMs

CALCULATE = 'calculate'

NULL_FAIL = 'null_fail'

ZERO_LEGACY = 'zero_legacy'

class orka.agents.local_cost_calculator.LocalCostCalculator(policy: str = 'calculate', electricity_rate_usd_per_kwh: float = None, hardware_cost_usd: float = None, hardware_lifespan_months: int = 36, gpu_tdp_watts: float = None, cpu_tdp_watts: float = None)

Bases: object

Calculate real operating costs for local LLM inference.

Cost components: 1. Electricity: GPU/CPU power consumption during inference 2. Hardware amortization: Depreciation of compute hardware 3. Cloud costs: If running on rented cloud infrastructure

__init__(policy: str = 'calculate', electricity_rate_usd_per_kwh: float = None, hardware_cost_usd: float = None, hardware_lifespan_months: int = 36, gpu_tdp_watts: float = None, cpu_tdp_watts: float = None)

Initialize cost calculator.

Parameters:

• policy – “calculate”, “null_fail”, or “zero_legacy”

• electricity_rate_usd_per_kwh – Local electricity rate (default: auto-detect)

• hardware_cost_usd – Total hardware cost for amortization

• hardware_lifespan_months – Hardware depreciation period

• gpu_tdp_watts – GPU power consumption (default: auto-detect)

• cpu_tdp_watts – CPU power consumption (default: auto-detect)

calculate_inference_cost(latency_ms: float, tokens: int, model: str, provider: str = 'ollama') → float | None

Calculate the real cost of local LLM inference.

Parameters:

• latency_ms – Inference time in milliseconds

• tokens – Total tokens processed

• model – Model name for optimization estimation

• provider – Local provider (ollama, lm_studio, etc.)

Returns:

Cost in USD, or None if null_fail policy

Raises:

ValueError – If null_fail policy is enabled

orka.agents.local_cost_calculator.get_cost_calculator() → LocalCostCalculator

Get the global cost calculator instance.

orka.agents.local_cost_calculator.calculate_local_llm_cost(latency_ms: float, tokens: int, model: str, provider: str = 'ollama') → float | None

Calculate local LLM inference cost.

Convenience function that uses the global calculator.

Returns:

Cost in USD, or None if null_fail policy

Raises:

ValueError – If null_fail policy is enabled