TrackCarbon logo | Fondation Sahar

Estimate the energy consumption of your AI usage with TrackCarbon

The massive deployment of artificial intelligence in many professions, sometimes abusively, is accompanied by agents that are very resource-intensive. The goal here is to estimate the energy consumption.

TrackCarbon dashboard with consumption charts
  • High granularity with personalized activity tracking
  • Simple and non-intrusive user experience
  • Clean visual interface with accessible metrics
  • Full transparency on estimation methods (open source code)
  • Adaptability and compatibility with different AI types
  • Modular architecture allowing customization according to needs

Context

In many situations, we use artificial intelligence (AI) models that are largely oversized compared to real needs, including in everyday tools where AI is now almost automatically integrated. Added to this is a lack of transparency: we often ignore the complexity of models and the resources mobilized behind the scenes, even to perform simple tasks.

Estimating the energy consumption and resources of AI appears essential today, especially in a context where its use is growing exponentially without real control. The goal is to build a common framework that makes AI use more responsible, without adopting an accusatory stance. It is about promoting a balanced approach that fosters innovation while ensuring sustainable and ethical use of AI technologies.

Presentation

The application aims to estimate the electrical cost and carbon footprint associated with the use of artificial intelligence. Installed on the user’s computer, it tracks the number of requests made by each agent, monitors them over time, and provides an energy estimate. We approach this challenge with an acute awareness of its complexity. The subject is vast and the proposed solution, although promising, is still in development and far from reaching its full potential. We approach this project with confidence, while being aware that its success will largely depend on community contribution. Active participation and constructive feedback are therefore essential to ensure the relevance and effectiveness of this evolving tool.

Call for Contribution

The project is part of an open source approach, with an estimation module open to contributions and freely accessible via the Foundation’s GitHub containing the necessary explanations to contribute to the project:

Foundation GitHub(opens in a new tab)

The estimation model is still in development: even though it is based on solid scientific publications, the task remains complex. The goal is to involve the community, and why not directly involve model providers, to better understand and refine these estimates.

How it works

The application counts and collects certain parameters present in the prompts and responses exchanged between the user and the AI. Based on research work and measured parameters, it calculates an estimate of energy consumption and the associated carbon footprint.

Compatibility

The application allows you to track and estimate the usage of different AIs. Here is the list of currently supported services:

ChatGPT Web
Claude Web
Claude Application
Gemini Web
Mistral Web

Warning: TrackCarbon is not yet compatible with the native ChatGPT application. Please exit TrackCarbon if you wish to use the ChatGPT application to avoid any malfunction. Compatibility is currently under development and will be available in an upcoming update. In the meantime, you can use ChatGPT without any problems via your web browser, where tracking works correctly.

Installation

When first launching the application, it will be necessary to follow a few installation steps, detailed below. The application is compatible with macOS Sequoia 15.3 and later.

Planned Evolutions

  • Granularity

    Increased granularity on tracked and displayed models allowing analysis customization

  • New Estimators

    Integration of new estimators developed by the community

  • Admin Interface

    To manage and control consumption of a fleet of users/computers

  • Report Generation

    To facilitate monitoring and control

  • Custom Dashboard

    With the ability to segment consumption by role within the organization

  • New Indicators

    Development of complementary indicators beyond carbon footprint

  • Accessibility

    Availability on all operating systems (Windows and Linux priority)

  • Query Routing

    Suggestion to route queries to the most suitable and less consuming model