alii13

AI agent usage 4% verified 4% of these tokens were counted and signed live on the author's Mac. The rest is self-reported history from before tokenproof was installed.

Verified since 27 Oct 2026. Last published 2 Nov 2026.
135.5B
Opus-equivalent tokens, all timeEvery token is weighted by what it would cost on Claude Opus 5.5, so a token from a cheaper model like Haiku counts for less. Pricier models count as one. This keeps a pile of cheap tokens from passing for real heavy use.
Last 30 days
135.7B raw tokensEvery token sent to and from the models, before price weighting. Most of it is context the agent re-reads from cache on each turn.393.3M generatedTokens the models wrote: code, answers and reasoning. The closest thing to how much work the AI produced. 108 active days0d streak · best 20d Claude 99.8%OpenAI 0.2%

Where the number comes from

5.9BVerified on-devicesigned live on the author's Mac since 27 Oct 2026tokenproof reads each response from Anthropic or OpenAI as it arrives and signs the token count with a key that never leaves the Mac's Secure Enclave chip. Apple certifies the key belongs to the genuine, unmodified tokenproof app, so the count can't be edited or faked.
129.6BSelf-reported historyimported once from local Claude Code and Codex logsUsage from before tokenproof was installed, read once from the log files Claude Code and Codex keep on the Mac and then frozen. Those files can be edited by their owner, so this part can't be proven. The consistency check below helps judge it.
History looks consistent with verified usetokenproof compares the average daily usage in the self-reported history with the average daily usage it has verified live on this Mac. Real history from a heavy user looks similar to their verified use; inflated history stands out as much higher. Here the history averages 1.2x the verified daily rate, within the 3x allowed.

Activity

Token activity heatmap

Models

Opus 5.5Claude32.5%
Opus 5Claude27.6%
Opus 4.8Claude23.7%
Fable 5Claude7.5%
Opus 4.7Claude3.9%
Fable 5.1Claude2.7%
14 more models
Opus 4.6Claude1.4%
Sonnet 5Claude0.2%
Haiku 4.5Claude0.1%
GPT-5.6 SolOpenAI<0.1%
GPT-6.1 SolOpenAI<0.1%
GPT-5.6 TerraOpenAI<0.1%
Haiku 4.5Claude<0.1%
Sonnet 4.6Claude<0.1%
Sonnet 4.5Claude<0.1%
Sonnet 5.5Claude<0.1%
Codex Auto ReviewOpenAI<0.1%
GPT-5.5OpenAI<0.1%
Opus 4.5Claude<0.1%
Haiku 5.5Claude<0.1%

Embed

Card[![tokenproof stats](https://tokenproof.pages.dev/alii13/card.svg "AI agent token usage from Claude Code and Codex, in Opus-equivalent tokens (cheaper models count for less). Verified = counted and signed live on the author's Mac via Apple App Attest, so it can't be edited. The rest is self-reported history from before tokenproof was installed. Click for the breakdown and to verify the signatures yourself.")](https://tokenproof.pages.dev/alii13)
Heatmap[![tokenproof activity](https://tokenproof.pages.dev/alii13/heatmap.svg "AI agent token usage from Claude Code and Codex, in Opus-equivalent tokens (cheaper models count for less). Verified = counted and signed live on the author's Mac via Apple App Attest, so it can't be edited. The rest is self-reported history from before tokenproof was installed. Click for the breakdown and to verify the signatures yourself.")](https://tokenproof.pages.dev/alii13)
Badge[![tokenproof](https://tokenproof.pages.dev/alii13/badge.svg "AI agent token usage from Claude Code and Codex, in Opus-equivalent tokens (cheaper models count for less). Verified = counted and signed live on the author's Mac via Apple App Attest, so it can't be edited. The rest is self-reported history from before tokenproof was installed. Click for the breakdown and to verify the signatures yourself.")](https://tokenproof.pages.dev/alii13)
Card, dark only[![tokenproof stats](https://tokenproof.pages.dev/alii13/card.svg?theme=dark "AI agent token usage from Claude Code and Codex, in Opus-equivalent tokens (cheaper models count for less). Verified = counted and signed live on the author's Mac via Apple App Attest, so it can't be edited. The rest is self-reported history from before tokenproof was installed. Click for the breakdown and to verify the signatures yourself.")](https://tokenproof.pages.dev/alii13)
Hovering an embedded widget shows: “AI agent token usage from Claude Code and Codex, in Opus-equivalent tokens (cheaper models count for less). Verified = counted and signed live on the author's Mac via Apple App Attest, so it can't be edited. The rest is self-reported history from before tokenproof was installed. Click for the breakdown and to verify the signatures yourself.”

Verify it yourself

Each Mac signs its numbers with a key locked in its security chip, and Apple certifies that key belongs to the genuine tokenproof app. The server re-checks every signature before publishing, and anyone can repeat the check with the files below.
rIvvZvAoG2fJaFDGkIXQHvxu+r4cgNudiFu3IYXYOL8=update #29280 · 2 NovEvery summary this Mac signs gets the next number. The server only accepts a higher number than it has, so an old summary can never replace a newer one.
Download the signed summaries and run verify/verify_ledger.py from the tokenproof repo.

Tokens are weighted by Claude Opus 5.5 prices, so cheaper models count for less and pricier ones count as one. Only daily totals per model are published: no prompts, no message ids, no times of day.