Track the Environmental Impact of Every AI Request

AI inference consumes real energy, and that energy carries a carbon and water cost that varies significantly depending on where and how a model is run. Meliai believes organisations should be able to measure, report, and minimise the environmental footprint of their AI usage — so every API response includes a detailed environment_impact block alongside the model output. This data is calculated per request based on the actual provider, datacenter location, and measured Power Usage Effectiveness (PUE), giving you an accurate picture rather than broad estimates. Whether you are building sustainability dashboards, meeting ESG reporting obligations, or simply trying to make greener infrastructure choices, this field gives you the raw data to act on.

The `environment_impact` Response Field

Every chat completion response, embedding response, and other inference response from Meliai includes an environment_impact object at the top level. Here is a representative example:

{
  "environment_impact": {
    "energy_kwh": 0.00015,
    "carbon_g_co2": 0.06,
    "water_liters": 0.0002,
    "renewable_percent": 85,
    "pue": 1.18,
    "provider_id": "ovhcloud",
    "location": "FR"
  }
}

number

Energy consumed by this request in kilowatt-hours, including overhead scaled by the datacenter PUE.

number

Carbon dioxide equivalent in grams, calculated from energy_kwh and the grid carbon intensity at the datacenter location.

number

Estimated water consumption in liters, derived from the datacenter’s Water Usage Effectiveness (WUE) and the energy consumed.

integer

Percentage of energy at the serving datacenter sourced from renewable generation, as reported by the provider.

number

Power Usage Effectiveness of the datacenter — the ratio of total facility energy to IT equipment energy. A PUE of 1.0 would be perfectly efficient; values closer to 1.0 are better.

string

The Meliai provider identifier that served this request, such as "ovhcloud", "scaleway", or "hetzner".

string

ISO 3166-1 alpha-2 country code of the datacenter that processed the request, for example "FR", "DE", or "FI".

Every Meliai inference response also includes a billing_cost object at the top level alongside environment_impact. This field carries the monetary cost of the request broken down by token usage, letting you track both financial and environmental spend in a single response.

Minimising Environmental Impact

Meliai’s :eco routing suffix biases the automatic provider selection toward datacenters with higher renewable energy percentages, lower grid carbon intensity, and better PUE scores. Switch to :eco by appending the suffix to your model ID — no other changes are required.

Python

response = client.chat.completions.create(
    model="<MODEL_ID>:eco",           # bias toward greener providers
    messages=[{"role": "user", "content": "Summarise this document."}],
)

impact = response.model_extra.get("environment_impact", {})
print(f"Carbon: {impact.get('carbon_g_co2')} g CO2")
print(f"Renewable: {impact.get('renewable_percent')}%")
print(f"Location: {impact.get('location')}")

Use :eco routing to prefer providers with higher renewable_percent and lower carbon_g_co2. Combined with locations such as Finland (FI) or Norway (NO) — where grids run on near-100% renewables — this can reduce per-request carbon by an order of magnitude compared to coal-heavy regions.

Logging and Aggregating Carbon Per Request

The example below iterates over a list of prompts, accumulates the carbon cost of each response, and prints a total. Adapt this pattern to write metrics to your observability stack, database, or ESG reporting tool.

Python

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["MELIAI_API_KEY"],
    base_url="https://api.meliai.ai/v1",
)

prompts = [
    "Summarise the Q3 earnings report.",
    "Translate this paragraph to German.",
    "List five marketing taglines for our product.",
]

total_carbon_g = 0.0
total_energy_kwh = 0.0

for prompt in prompts:
    response = client.chat.completions.create(
        model="<MODEL_ID>:eco",
        messages=[{"role": "user", "content": prompt}],
    )
    impact = response.model_extra.get("environment_impact", {})
    carbon = impact.get("carbon_g_co2", 0)
    energy = impact.get("energy_kwh", 0)
    total_carbon_g += carbon
    total_energy_kwh += energy
    print(f"[{impact.get('location', '?')} / {impact.get('provider_id', '?')}] "
          f"{carbon:.4f} g CO2 — {energy:.6f} kWh")

print(f"\nTotal carbon:  {total_carbon_g:.4f} g CO2")
print(f"Total energy:  {total_energy_kwh:.6f} kWh")

response.model_extra is the OpenAI Python SDK’s way of accessing fields that are not part of the standard OpenAI response schema. If you are using a different HTTP client, environment_impact is present at the top level of the JSON response body alongside id, object, choices, and usage.

Overview

Models & Routing

Features

Privacy

Track the Environmental Impact of Every AI Request

The `environment_impact` Response Field

Minimising Environmental Impact

Logging and Aggregating Carbon Per Request

​The environment_impact Response Field

​Minimising Environmental Impact

​Logging and Aggregating Carbon Per Request

The `environment_impact` Response Field

Minimising Environmental Impact

Logging and Aggregating Carbon Per Request