BETA RELEASE
Skip to main content

Responsible AI Chatbots

“Green AI” gets thrown around a lot. This page lays out what each product actually offers, side-by-side, so you can compare. If you spot an error or omission, let us know.

FULL DISCLOSURE: Dangerous Robot is sponsored by TreadLightlyAI.

Summary

Quick-glance summary of each product's stance on AI ethics, financial transparency, environmental impact, and other notes.
Product AI ethicsEnvironmentalFinancial transparencyNotes
GreenPT Open-weight models only. Hosted at Scaleway data center in France (PUE 1.37), powered by 100% renewable energy. GreenPT BV, Netherlands; no published financials. Smaller, more efficient models.
Euria Open-weight models (US, EU, CN). Infomaniak data center (PUE 1.06, waste heat to district heating). 100% renewable energy. Receipts Infomaniak is a Swiss B Corp; no published financials. Bold environmental committments. Certified by myClimate.org.
Ecosia AI Proprietary model (US), GPT-4 mini. Conventional cloud hosting; funds tree planting projects worldwide Non-profit steward-ownership; publishes financial reports. Free, ad-supported.
TreadLightlyAI Proprietary models (US) and open-weight models (US, FR). Infomaniak (PUE 1.06, waste heat to district heating) + Anthropic conventional cloud. Sole proprietorship; no published financials. Beta product from same founder as Dangerous Robot.
ChatGPTree Proprietary model (US), GPT-4 Conventional cloud hosting; funds tree planting via Evertreen and Veritree. No published financials. Evertreen partner flagged for non-transparent practices, Veritree is transparency-focused.
Earthly Insight Proprietary models (US). Conventional cloud hosting. 33% of revenue to Global Rewilding Alliance No published financials. Launched October 2025.
Viro AI Proprietary models (US) and open weight models (US, CN). Conventional cloud hosting; offsets via Terrapass RECs; methodology published. Solo-founder LLC; Viro Climate Action, Inc; no public disclosure. Partner Terrapass entered a California DFPI consent order in July 2025.

Comparison table

Filter
Products
Feature GreenPT Euria Ecosia AI TreadLightlyAI ChatGPTree Earthly Insight Viro AI
Environmental
Hosted on renewable energy Yes Yes No Partial All but Anthropic No No No
Real-time energy display Yes No No Yes No No Yes
No image generation Yes Yes Yes Yes Yes Yes No
Models & AI safety
Models Mistral Small, GPT-OSS DeepSeek R1, Qwen 3, Mistral Nemo, Llama 3 GPT mini Mistral Ministral & Small, GPT-OSS, Claude Haiku & Sonnet GPT GPT, Gemini Flash & Pro, Claude Sonnet GPT, Claude, Gemini, LLaMA
Open-source models Yes Yes No Partial Claude is proprietary No No Partial
Product Features
Image and document analysis Yes Yes No Yes No Yes Yes
Agent Web search Yes Yes Yes Yes No No No
Normal Web search No No Yes Yes No No Yes
Accessibility Yes Level A Partial Yes Level AA+ Yes Level AA+ Partial Yes Level A Partial
Privacy & data
Data used for training No (negative) No (negative) No (negative) No (negative) Unknown probably? Unknown probably not? No (negative) Ads based on chat content
Data jurisdiction Swiss (FADP + GDPR) Swiss (FADP + GDPR) German (GDPR) US US US US
Business & cost
Published financials No No Yes No No No No
Ownership structure Private B Corp Non-profit, Steward-ownership Sole proprietorship Private Bootstrapped LLC Solo-founder LLC
Free tier Yes Yes includes kSuite Yes Ad supported Yes Yes Yes Yes Ad supported
Pricing € 4.50 | 17.50 /mo € 1.58 /mo includes kSuite Pro n/a $ 3 | 8 | 20 /mo $ 11 /mo $ 20 /mo $ 1 | 10 /mo
Maturity 2 years 6 months 2 years (AI feature) 2 months (Beta) 1 year 9 months 9 months

Data gathered from public sources and direct research. If something is wrong, let us know. Last updated: .

Legend:
  • Yes
  • No
  • No (preferred)
  • Partial
  • Unknown

Footnotes

  1. Ecosia energy displayShows a tree counter (how many trees your searches have funded) rather than per-query energy data. This is an engagement metric, not energy transparency.
  2. Euria Chinese-origin modelsDeepSeek R1 and Qwen 3 have documented censorship of topics politically sensitive to the CCP. Running open-weight versions on Swiss infrastructure avoids sending data to China, but training biases are baked into the model weights.
  3. GreenPT green claimsCover inference-time energy only. Mistral models were trained on conventional infrastructure. This tension between green hosting and non-green training applies across the industry.