Skip to main content

Sustainability

AI is energy-hungry. Here is how we think about that, and what we are doing about it.

Energy and infrastructure

We run on Cloudflare Workers, which uses a globally distributed network with strong efficiency metrics. We choose model providers in part on their published energy commitments and prefer smaller, task-fit models over the largest available where the quality is acceptable.

Reducing waste

We cache prompts where it makes sense, batch requests where latency permits, and surface usage transparently to builders so they can right-size their assistants.

Where we want to go

Publishing per-conversation energy estimates, choosing model providers on their grid mix, and giving builders tools to set energy budgets for their assistants.