CodeCarbon Local compute emissions

EcoLogits GenAI API calls emissions

Website · CodeCarbon docs · EcoLogits docs · Discord

Code Carbon is a French non-profit (NGO). We build free, open-source software to help developers and researchers understand the energy and carbon impact of their work.

Most of what you see here is maintained by volunteers — researchers, engineers, and sustainability practitioners who contribute in their spare time. Sponsors and partner organisations provide hosting, funding, and expertise; the day-to-day work still runs on community effort.

People also build integrations outside these repos — editor plugins, status bars, and tooling that connect to our libraries and APIs. We highlight the ones we know about below; if you ship something similar, tell us on Discord.

AI and software need electricity. Where that code runs — on your laptop or on a provider's servers — changes what you can measure directly.

	CodeCarbon	EcoLogits
What it tracks	Emissions from local computing — code on hardware you control	Emissions from GenAI API calls — OpenAI, Anthropic, Mistral, etc.
Typical use cases	Model training, local inference, scripts on your machine or cloud VM	Chatbots, agents, RAG pipelines calling remote models
How it works	Measures CPU, GPU & RAM power, applies regional grid carbon intensity	Estimates impacts from request metadata (model, tokens, latency) via LCA models
Get started	pip install codecarbon	pip install ecologits

They are complementary, not competing. CodeCarbon is for compute you run yourself — training, local inference, jobs on your machines or cloud VMs. EcoLogits is for when you call third-party GenAI APIs and want to estimate the impact of those requests.

When to use CodeCarbon vs EcoLogits →

Track CO₂ from code running on your own hardware. Python library, CLI, cloud dashboard, and offline visualisation.

from codecarbon import EmissionsTracker

tracker = EmissionsTracker()
tracker.start()
# your code
emissions = tracker.stop()

→ Documentation · Dashboard · Website

Python library that attaches environmental impact estimates to responses from official provider SDKs.

from ecologits import EcoLogits
from openai import OpenAI

EcoLogits.init(providers=["openai"])
response = OpenAI().chat.completions.create(model="gpt-4o-mini", messages=[...])
print(response.impacts.gwp.value.mean, "kgCO2eq")

→ Documentation · Methodology

Related EcoLogits projects:

These projects are maintained outside the mlco2 organisation but built on our open APIs and methodology. They are independent — not official releases — and we are grateful to the authors for extending the ecosystem.

_{VS Code status bar integration by marmelab/ecologits-vscode}

Built something on CodeCarbon or EcoLogits? Open a PR to mlco2/.github or ping us on Discord — happy to list it here.

We welcome issues, pull requests, documentation fixes, and methodology feedback — whether you have an hour or a recurring slot.

• CodeCarbon: Contributing guide
• EcoLogits: Contributing guide
• Chat: Discord
• Support the NGO: HelloAsso — Code Carbon

Our tools stay free thanks to organisations that sponsor infrastructure, funding, or expertise. Community volunteers do the rest.

Infrastructure & core sponsors (CodeCarbon)

           

EcoLogits sponsors

Historical & research partners: Mila, BCG GAMMA, Comet, Haverford College, Boavizta

_{Made with care by volunteers at Code Carbon · mlco2 on GitHub}