The line is going up.

More machine activity. Let’s see what it can do.

Sep 1, 2026 1,436,197 output tokens / day · 30-day average

Inspect: hover, tap, or ← →. Esc resets. Home selects the first date; End selects the latest.

How this is counted

Trailing 30-calendar-day averages from retained native Codex and Claude logs. Output is recorded output, including recorded Codex reasoning tokens. Provider accounting differs, and missing local records are not proof of zero use elsewhere. These counts measure activity, not quality.

Input separates cache reads from non-cache input, including cache-creation writes. Input and output are different measurements and are never added together.

The public chart refreshes from aggregate-only local history when Johnny's Mac is available. If a refresh fails, the last verified dataset remains in place.

Token counts may understate the growth in my data-center demand. If the models I use get larger, even a flat line can mean more compute and GPU memory used over time. A model-size adjustment may eventually be possible; for now, the model and serving details aren’t visible enough to estimate it reliably.

The questions should
get harder.

Can

Johnny’s
AI

run a (Turo) business?

Market watching. Fleet economics. Johnny makes the calls.

Turo · live experiment Open

manage an investment portfolio?

Investment theses, tested against the evidence.

Stonk · public cockpit Open

prepare you for a doctor visit?

Your medical record, turned into better questions.

Pearl · patient preparation Open

understand a restaurant?

A clearer picture of daily operations.

Redbo · operating system Open

find a market hiding in plain sight?

A market map built from scattered evidence.

Refugee settlement · market map Open

do his taxes?

A public Schedule C, transaction by transaction.

Taxes · public decision ledger Coming soon