Sustainability
Mitigate your carbon footprint with on-device models and cloud models powered by 100% renewable energy. Track it all on our sustainability dashboard.
An Environmental Crisis
AI data centers can consume as much power and water as entire cities. When new data centers come on line, communities are flooded with noise pollution and emissions from gas generators. We can change that.
Weave helps you mitigate your impact
Track the energy used to generate responses and analyze data using AI.
Efficiency and Sustainability scores help you make more informed decisions about usage.
Sustainability Dashboard
Use the Sustainability Dashboard to track important metrics related to your usage over time.
Methodology
Our Methodology
Overview & Guiding Principals
Purpose & Scope: Weave is committed to tracking and reducing the environmental impact of AI usage. Our methodology focuses on measuring the operational energy and CO₂e emissions associated with AI inference.
Weave's Commitment:
Real Data First: We prioritize actual measured data over estimates.
Conservative Estimation: When direct data is unavailable, we err towards higher estimates.
Transparency: We disclose all factors and data sources, as can be accessed through the Full Methodology Report.
Privacy Preservation: Our calculations are performed offline or in our secure backend to ensure user privacy.
Understanding Your Dashboard Metrics
Sustainability Score: Measures the proportion of your AI usage that employs eco-friendly models.
Efficiency Score: Measures the energy efficiency per token across use of different models.
CO₂e Saved: Calculated based on a counterfactual baseline (e.g., all usage through non-eco friendly cloud provider).
Energy Used & CO₂e Emitted: Total energy consumption and CO₂e emissions from your AI usage.
Environmental Equivalents: We translate your impact into tangible metrics, such as "Kilometers Not Driven" and "Tree-Days of Absorption."
How We Measure Your Impact
On-Device Inference:
Energy Consumption: We use real, hardware-measured energy consumption data for MLX/Ollama, utilizing CPU, GPU, and DRAM counters.
Regional Grid Carbon Intensity: We employ a privacy-preserving, offline method to detect and apply regional grid carbon intensity data.
Cloud Inference - GreenPT:
Energy and Emissions Data: GreenPT provides provider-reported energy and emissions data, sourced from 100% renewable energy.
Industry-Leading Metrics: GreenPT's data centers boast a Power Usage Effectiveness (PUE) of 1.37 and Water Usage Effectiveness (WUE) of 0.067, significantly below industry averages.
Cloud Inference - Other Providers (Google, Anthropic, OpenAI):
Informed Estimates: Due to a lack of publicly available data from these providers, we use informed estimates.
Token-Based Estimation Model: Our model weighs input versus output tokens to estimate energy and emissions.
Transparency & Limitations
Known Limitations and Future Work:
Estimation Uncertainty: Significant uncertainty exists for cloud providers with their limited data transparency.
Annual Average Grid Intensity: We currently use annual averages rather than real-time grid conditions.
Scope Exclusions: Our methodology does not include training emissions or embodied carbon.
Future Improvements: We are actively working to engage with providers, explore real-time grid APIs, and enhance our methodology.
OUR MISSION & VISION
Committed to providing eco-friendly AI through efficient software and fantastic partnerships.
Software
Weave lets you see your AI usage over time and a transparent estimate of environmental impact. Where your device and cloud models expose measurable signals (like runtime, hardware type, and token usage), we use them directly; where they don’t (like data-center water use), we use published datasets and clearly stated assumptions to estimate energy, CO₂e, and water impact. All reporting is designed to preserve privacy: your content stays on your device, and we only process the minimum metadata needed to produce the report.
Weave also makes it easy to run AI on your own device with no setup. For many everyday tasks, smaller, efficient models can significantly reduce energy use compared to large, remote models, and Weave is built to default to those options when they meet your needs. Available for Free with Basic.
Partnerships
Big solutions require great partners, which is why we have partnered with GreenPT, a cloud AI provider who hosts AI models powered by 100% renewable energy.
Partnering with GreenPT allows us to mix together the efficiency of on device AI with the power of larger cloud AI. GreenPT data centers have a PUE of 1.37 and WUE of 0.067, far below the industry average, and their commitment to 100% green power makes them the perfect partner and provider for our cloud solution. Available with Eco-Pro.