Better tools should come with greater responsibility. Our direction is simple: use AI deliberately, reduce avoidable computation, and support environmental efforts beyond our product.
Explore our approach2%
We commit to donating 2% of all revenue to environmental impact efforts.
This is a commitment, not a report of completed donations. Recipient details and donation records have not yet been published.
Concerns behind “quit big AI” and wider AI criticism deserve a serious response: energy and water use, infrastructure demands, and whether every generation creates real value. We want to earn trust through useful choices and clear evidence, not green labels.
Efficiency direction
Our goal is to use smaller, task-appropriate models for simpler work and reserve more capable models for tasks that genuinely need them. Existing provider routing supports configurable model choices; it is not a verified lowest-emissions routing system.
Evidence standard
We aim to publish usage boundaries, estimation methods, and provider data as they become available. We do not currently publish verified energy, water, or carbon measurements, and a smaller or cheaper model is not automatically cleaner.
Future goals
Our priorities are fewer unnecessary generations, more reuse of useful outputs, and better visibility into the impact of text, voice, image, and video workloads. These are improvement goals, not measured reductions.
Accountability pipeline
Explore how we intend to reduce, measure and account for generative AI use. These stages describe a proposed operating process, not a live emissions tracker, completed offset program or automated donation system.
Smaller models are not automatically lower-emissions. Environmental giving is not the same as verified carbon compensation.
Limit duplicate generations, retries and idle sessions. Route straightforward tasks to suitable smaller models when quality and availability permit.
Model routing and request logs
Capture provider, model, modality, token totals, generation time and relevant billing metadata across text, voice, image and video.
Provider event logs and usage ledger
Calculate bounded estimates only where published provider factors or reliable energy data exist. Disclose unknowns instead of fabricating precision.
Versioned methodology and coverage
Reconcile recognized revenue and earmark 2% for eligible impact initiatives, with adjustments for refunds and reporting periods.
Revenue reconciliation and allocation ledger
Record recipient, payment date, amount, receipt and purpose for each verified contribution. Community investments and environmental projects are reported separately.
Receipts and recipient verification
Publish actual contributions, estimated footprint and limitations. Only label emissions as offset after independent verification and retirement evidence.
Public report and retired-credit references
