keymail is AI-written, and the honest carbon ledger has two halves: what it cost to build it (the AI tokens spent writing it) and what it costs to run it (the whole fleet, every day). Building is one-time and grows only as more AI development happens; running is ongoing.
Last measured
Token counts are exact — Claude Code records per-session usage locally, and tools/carbon-tally.mjs accrues them into the committed ledger this page reads. The carbon conversion is an estimate with wide error bars, because no model provider publishes per-token figures: we band public calibration points and always show the range, never just the flattering end.
The run estimate is arithmetic on the fleet's real shape — two ARM boxes (t4g.nano), their banded wall-power share, grid intensity for eu-west-1 (Ireland), and datacenter overhead (PUE) — committed as a daily rate and multiplied out by your browser at render time. Known undercounts are listed at the bottom of the Built section, on purpose.
Couldn’t load the carbon tally. It may be mid-publish — this page retries every minute.
A running estimate of the CO₂e cost of the LLM tokens spent building keymail. Token counts are exact; the conversion is banded — —. This grows only when more AI development happens.
That’s about the same CO₂e as, centrally:
The CO₂e cost of running the whole keymail fleet — today, two tiny ARM boxes terminating their own TLS — banded low to high. The daily rate is committed with the ledger; the cumulative figure is computed live from the day the fleet first went up.
Cumulatively so far, that’s about the same CO₂e as, centrally:
Operational estimate pending — the committed tally doesn’t carry a run block yet. Nothing here is ever shown as a fake zero.