Trust / What we track
What we track
Every way usage gets into Centrail — live, in progress, and what we're fighting for.
Centrail's promise: whichever way you use AI, we crunch the best numbers available for it — and we're public about which is which. We think AI is worth keeping — and keeping it starts with honest measurement.
Coverage today
| Source | How | Quality | Status |
|---|---|---|---|
| AI coding tools (Claude Code, Copilot CLI, Codex) | npx centrail connect — local log capture | Exact tokens · Verified | Live |
| ChatGPT (chat app) | Data export → browser-side import | Estimated · Measured | Live |
| Claude (chat app) | Data export → browser-side import | Estimated · Measured | Live |
| Gemini (chat app) | Google Takeout → browser-side import | Estimated · Measured | Live |
| Quick estimator (no data needed) | The calculator — pick models, type token amounts | Rough · Estimated | Live |
| Org API usage (OpenAI, Anthropic) | Previously connected admin keys still pull; new admin-key connections are retired for security | Exact tokens + real billed cost | Existing connections only |
| Image generation | Counted from chat exports + org APIs | Estimated | Planned |
| CI agents (GitHub Actions) | CLI in headless mode | Exact tokens | Roadmap |
| Enterprise chat workspaces | Provider admin APIs | Exact | Roadmap |
What we're fighting for
None of the major AI providers publish per-user energy, water, or carbon numbers. Until they do, every figure here is an estimate built on cited public research — see the methodology. We believe the companies building these models should publish the real numbers, and every person who tracks their footprint with Centrail makes that ask louder.
Want to help, or want a source added? Email support@centrail.org — this page is the roadmap.