Launching today
CrbonFree
Audit-grade carbon numbers for every AI token you use
7 followers
Audit-grade carbon numbers for every AI token you use
7 followers
How much energy and carbon does your AI use? CrbonFree meters every call, per model and per provider, and turns the answer into a report an auditor can check.
CrbonFree
Hi everyone, Sadi from Crbon Labs.
We built CrbonFree because AI had become a real line in company budgets, but it was missing from their carbon inventories. Every tool we found estimated it, usually from a calculator or an average per query. An auditor can't sign off on a guess, and from September 2026 the EU bans the carbon-neutral claims that used to paper over the gap.
CrbonFree connects to the providers you already pay (OpenAI, Anthropic, OpenRouter, Bedrock, Vertex and the Vercel AI Gateway) and reads the usage they meter, token by token. Each call gets an energy, water and CO2e figure. We publish the whole methodology, down to every factor and formula, and each figure comes with an uncertainty band. You end up with a report that someone outside your company can check.
Measurement is the centre of the product. Climate action is separate and optional: if a customer retires credits, that money never changes the measured footprint. We also put the numbers where engineers already work, so there's a CLI, TypeScript and Python SDKs, an MCP server, and a browser extension for the ChatGPT and Claude use that never touches an API key.
Our first beta customer is a billion-token-a-month user, and the SDKs and MCP server have passed 4,500 downloads with zero paid marketing. We write up what we measure on the blog, most recently what Meta's 73.7 trillion tokens a month weigh per employee.
Ask us anything about the methodology. It's the part we're proudest of, so try to break it.