AI usage by day, source, and work driver.

Exact local logs for Codex and Claude Code, labelled estimates for local models — bucketed by London day and pointed at one question: what should the computer do next?

Data through 2026-09-12 · last extracted 2026-09-12

Claude Code turns131,841assistant turns

Active days121rows in view

Measured non-cache floor1.9BCodex split unavailable

Context processed33.4Bincl. cached re-reads

Daily burn

Heatmap

Weekly trend

Log-scaled trend

Source split

Exact beside estimated

exact1.3BCodex / 4%

exact30.4BClaude Code / 91%

exact645.7KPerplexity Sonar / 0%

estimated1.8BLocal · MLX / 5%

estimated1.7KLocal · Ollama / 0%

Read Claude Code as context throughput, not spend. Its observed uncached input plus generated output is 0.5% (137.1M); the remainder is cache creation and cache reads. MLX reports prompt tokens only. Codex composition is unavailable, so its processed total is excluded from the non-cache floor.

Drivers

What is burning tokens

research57%

review36%

shipping8%

planning0%

Claude Code by estate

Foundry (harness)90%

KF (Knowledge Forge)8%

Council2%

Other1%

Scale equivalents

Make the number human

Measured non-cache floor — excludes Codex

Approx. words1.4Btokens × 0.75 words

Reading time95,587 h250 words per minute

Novel equivalents15,93190k words per novel

If every token were a written word

1.4B words

1.9B measured non-cache tokens × 0.75 · excludes Codex

That is 15,931 novels — each book below = 266 novels

≈ 2,443× the length of War and Peace

Tolstoy’s novel runs to about 587,000 words — the canonical doorstop.

1.4B words ÷ 90,000 words/novel = 15,931 novels · 2,867,608 printed pages (÷ 500 words/page) · ÷ 587,000 words (War and Peace) = 2,443×

Reading it aloud at 250 words/min

Your token burn, read aloud95,587 h

A 40-hour work-week40 h

A 2,000-hour working year2,000 h

1.4B words ÷ 250 wpm ÷ 60 = 95,587 h · 2,390 work-weeks · 48 working years · 11,948 eight-hour days

Day detail · peak day

2026-09-11

research

Total burn2.7B

Claude Code2.5B9,390 responses · 0.5% uncached input + output

Codex170.5M

Local · MLX51.4M86,899 req · floor

Local · Ollama0

Estate (CC)KF 0% · Foundry 100% · Other 0%

Project-level detail (commits, top projects) loads only when running locally.

Moving-average table

Last 30 days

Click a row for that day’s detail.

Date	Total · M	7d avg · M	Codexexact · M	Claude Codeexact · M	CC calls	Sonarexact · K	MLXest · M	Driver
2026-09-12	44.1	1,019.8	12.7	4.9	38	0.0	26.5	research
2026-09-11	2,739.3	1,118.5	170.5	2,517.4	9,390	0.0	51.4	research
2026-09-10	1,709.1	824.6	68.4	1,589.9	8,403	0.0	50.7	research
2026-09-09	566.6	676.1	96.8	409.0	2,429	0.0	60.8	research
2026-09-08	321.1	647.0	10.1	253.0	1,697	0.0	58.0	research
2026-09-07	911.2	747.8	19.6	834.5	5,241	0.0	57.1	review
2026-09-06	847.0	810.6	5.9	775.6	5,183	0.0	65.5	research
2026-09-05	735.4	763.1	1.6	679.4	3,159	0.0	54.4	research
2026-09-04	681.7	717.6	1.0	634.6	1,846	0.0	46.2	research
2026-09-03	670.0	675.3	18.5	601.4	2,122	0.0	50.2	research
2026-09-02	362.3	622.3	1.1	298.9	1,279	0.0	62.3	research
2026-09-01	1,027.3	587.0	4.3	968.1	4,344	0.0	55.0	review
2026-08-31	1,350.4	461.4	71.6	1,222.6	3,691	0.0	56.2	research
2026-08-30	514.3	298.5	49.2	410.2	1,872	0.0	55.0	research
2026-08-29	417.1	325.5	3.3	354.2	1,052	0.0	59.6	research
2026-08-28	385.5	462.5	10.1	326.2	1,432	0.0	49.2	review
2026-08-27	298.8	531.9	0.0	248.7	1,351	0.0	50.1	research
2026-08-26	115.5	580.1	4.6	60.2	505	0.0	50.7	research
2026-08-25	148.4	630.9	0.3	97.5	652	0.0	50.6	research
2026-08-24	209.8	662.1	0.0	156.6	845	0.0	53.2	research
2026-08-23	703.7	719.5	103.5	543.1	2,401	0.0	57.1	review
2026-08-22	1,376.1	745.9	30.7	1,288.5	5,423	0.0	56.9	shipping
2026-08-21	871.3	630.6	27.1	788.4	4,674	0.0	55.8	review
2026-08-20	636.3	575.3	73.5	512.2	2,575	0.0	50.6	research
2026-08-19	470.5	584.3	8.8	414.6	2,623	0.0	47.1	review
2026-08-18	366.8	554.2	14.0	303.6	1,194	0.0	49.1	research
2026-08-17	612.1	595.0	3.9	552.4	2,276	0.0	55.8	research
2026-08-16	888.6	604.3	24.3	808.5	3,368	0.0	55.8	research
2026-08-15	568.8	563.8	9.7	504.1	1,971	0.0	55.1	research
2026-08-14	484.0	525.0	1.3	435.9	1,757	2.5	46.8	research

Method & fidelity

How each lane is counted

Days are bucketed in Europe/London. Exact lanes come from real local logs; estimated lanes are labelled and never presented as exact.

exact
Codex — sum of observed last_token_usage.total_tokens from ~/.codex/sessions/**/rollout-*.jsonl, suppressing repeated cumulative snapshots and identical events copied into forked rollouts. This is processed-token usage; historic cache and output composition may be unavailable.
exact
Claude Code — sum of input + cache_creation + cache_read + output once per model response from ~/.claude/projects/**/*.jsonl, identified by message ID with UUID fallback. Repeated content blocks count once. CC calls = unique model responses with usage.
exact
Perplexity Sonar — API-reported prompt + completion tokens per live-search call, from the local sonar-search CLI’s usage log. Powers weekly market-signal diffs and ad-hoc research.
est
Local · MLX — prompt tokens measured from the mlx_lm.server log (progress: x/Y ⇒ Y prompt tokens per request). Output tokens are not logged, so this is a conservative prompt-token floor; no output multiplier is assumed.
est
Local · Ollama — ≈ message characters ÷ 4 from the Ollama app database; Ollama does not record token counts.

The charts read scrubbed daily totals from data/daily-burn.json (committed). Day-detail (commit counts, project names) is read server-side from the git-ignored private/ file via /api/detail — never bundled into the page or committed, so a deployed build would show none of it. Re-run npm run extract to add a day.