Sustainability of AI Tools & Models
Last updated: 18 August 2026
Kortyx is an AI chief of staff for sales leaders. The environmental footprint of the AI tools and models we use is something we take seriously and design against. This page documents the considerations made, the choices in our architecture, and the commitments of the sub-processors that power the Service.
A signed PDF version of this statement is available for procurement reviews: Download Kortyx Sustainability Statement (PDF).
1. Architectural choices that reduce footprint
- Tiered model routing. The majority of Kortyx requests are routed to smaller, more efficient models (e.g. Gemini Flash / Flash-Lite, GPT-5 mini / nano). Larger frontier-tier models are used only when the task genuinely requires them (deep reasoning, long-context briefings). This materially reduces compute per interaction versus a "always use the largest model" approach.
- No model training on customer data. Kortyx does not train or fine-tune any model on customer content. We do not run training jobs — the most energy-intensive class of AI workload — at all.
- Data minimisation. We retain only what the Service requires, and offer export & deletion on request. Less stored data means less ongoing storage, replication and backup energy.
- Serverless, shared infrastructure. Kortyx runs on shared edge and serverless infrastructure rather than dedicated always-on servers, improving utilisation.
- No on-prem GPU fleet. Kortyx does not operate its own GPU datacentre. Inference is delegated to hyperscale providers whose datacentres have substantially better PUE (Power Usage Effectiveness) than typical enterprise datacentres.
2. Sub-processor sustainability commitments
Kortyx's AI inference is delivered through the sub-processors listed on our Sub-processors page. Their publicly stated sustainability positions, which Kortyx relies on as part of its processing chain:
| Provider | Stated commitment |
|---|---|
| Google (Gemini models, via Google Cloud) | Carbon-neutral operations since 2007; targeting 24/7 carbon-free energy across all datacentres by 2030. Google Cloud regions publish carbon characteristics per region. |
| OpenAI (via Microsoft Azure) | OpenAI inference runs on Microsoft Azure, which is committed to being carbon negative, water positive and zero waste by 2030, and to matching 100% of electricity consumption with zero-carbon energy purchases. |
| Lovable (hosting & AI Gateway) | Application hosting on shared cloud edge infrastructure (Cloudflare / equivalent), which operates on 100% renewable energy. |
3. What we do not currently do
In the interest of being straightforward with procurement teams:
- Kortyx does not publish a per-request gCO₂e figure. The underlying model providers do not yet expose this at the API level.
- Kortyx does not currently purchase separate carbon offsets — we rely on the renewable-energy and carbon commitments of the underlying infrastructure providers above.
- Kortyx does not operate its own datacentre, and therefore does not publish a PUE figure of its own.
We will update this statement as model providers expose more granular emissions data.
4. Engagement & documentation
Sustainability of the underlying tools and models is referenced during vendor engagements and documented here, on the public Sub-processors page, and in our Security & Data Protection statement. Enterprise customers may request a copy of the signed PDF statement linked at the top of this page as part of their vendor onboarding pack.
5. Contact
Questions about Kortyx's sustainability position, or requests for additional documentation for a vendor review, can be sent to hello@kortyx.ai.