Carbon Tracking & Financial Governance

Bauxite Intercept provides real-time visibility into the environmental and financial impact of your AI operations. By combining model telemetry with regional grid data, Bauxite enables precise departmental chargebacks and sustainability reporting.

How It Works

1. Energy & Carbon Estimation

Bauxite tracks the duration and token count of each request. It maps your backends to geographic regions (e.g., us-east-1 vs. eu-west-1) and applies Grid Carbon Intensity (gCO2eq/kWh) factors.

2. Granular Policy Buckets

Every request is tagged with a Policy Label. This allows enterprises to break down emissions and costs by internal department or project.

Example Prometheus Metric:

# Tracking CO2 by policy
bauxite_carbon_grams_total{policy="marketing-team"} 12.45
bauxite_carbon_grams_total{policy="rd-department"} 450.12

3. Financial Chargebacks

Bauxite calculates the USD cost of every prompt based on the specific provider’s pricing model (input vs. output tokens). These costs are aggregated per policy, enabling finance teams to accurately allocate LLM spend.

Configuration

Define your regions and model costs in config.yaml:

routing:
  backends:
    - url: "http://mock-provider:9091"
      provider: "openai"
      region: "europe-west" # Higher renewable mix

providers:
  openai:
    models:
      "gpt-4":
        input_cost_1k: 0.01
        output_cost_1k: 0.03
        energy_factor: 1.0

“None” Fallback

If a request does not match any specific policy, it is automatically bucketed under the none label. This ensures that 100% of your AI traffic is accounted for in sustainability and financial audits.

Reporting

Metrics are exported via the /metrics endpoint in standard Prometheus format, ready for visualization in Grafana.