A better world through efficient AI.

The short answer: same capabilities, more transparency. Your data stays in your region, and every conversation shows you the actual energy and carbon it consumed, measured from the hardware running it and the grid powering that hardware. Not estimates. Not annual averages. The real number.

Plans start at $4.99/month for individuals, $18.99/month for professionals and teams. Every plan includes a 14-day free trial. No credit card required.

No. If your team has used ChatGPT or any chat-based AI, they can use GreenPT immediately. Our Apps feature even removes the need for prompting skills for specific tasks. You can also import your existing chat history from ChatGPT, so everything you have built up carries over.

For US users, data is hosted in North America. Your prompts and responses are processed in real-time and not retained beyond short-lived operational caching. For users in other regions, data is hosted regionally. We don't train models on your inputs, wherever you are.

• GreenL (Mistral Small 3.2, 24B): fast everyday chat, with Dutch language fine-tuning.
• GreenR (GPT-OSS 120B): advanced reasoning, vision, and document analysis.
• Gemma4 (Google Gemma 4, 31B): multimodal, with a 256k context window.
• Plus additional open-source models available via our AI router.

We host only models locally on our servers. We don't use any external APIs or services. This means your data never leaves our infrastructure and is not shared with third parties. The models themselves don't store any conversation data.

Our models are carefully selected and fine-tuned to deliver performance comparable to proprietary alternatives while being more environmentally conscious and cost-effective.

We implement strict data protection measures and fully comply with GDPR requirements. All data is processed in North America, and we maintain transparent data handling practices. Our self-hosted model approach means your data never leaves our secure infrastructure.

Every request's energy is measured directly from GPU hardware, not estimated. Every carbon number uses the live carbon intensity of the grid running your workload. No annual averages. No projections. We report the physical carbon intensity of the grid our servers actually run on, not the carbon we could claim by buying renewable energy certificates elsewhere. The atmosphere doesn't distinguish between a certificate and a smokestack. Neither do we.

We acknowledge that the initial training of AI models, including those we use, wasn't performed on green infrastructure. However, we're working with the open-source community to develop more sustainable training practices for future models.

• Measured infrastructure. Energy and carbon measured on every request.
• Efficient model selection. The right model for the job.
• Technical optimisations. Smaller, efficient models.

Modern GPUs report their power consumption in real time. For every request, we record the GPU's energy counters when processing starts, track power throughout, and calculate the actual energy consumed. We then apply the current carbon intensity of the grid where that GPU is located, using live data from grid operators. That gives a real per-request carbon number, not a projection or an average. Most AI providers don't do this. If you bill per token, hardware-level measurement isn't in your interest. We think users deserve to know what their AI actually costs.

Because we can't honestly claim it. Not every grid our US infrastructure touches is fully renewable, and buying renewable energy certificates to paper over that doesn't change what actually flows into our servers. Instead, we tell you the truth: the real carbon intensity of the grid your request ran on. Some regions are very low-carbon, others less so. We show you the actual number, per request. That's more useful than a blanket "we're green" claim.