The numbers

Everything on this page is a number a stranger can check. Each line names the figure, the day it was measured, and exactly how, so you can run the same check yourself and get the same answer. Nothing here is a projection or an estimate dressed up as a fact.

Commits

42,400 unique git commits in the last 12 months, across 58 local repositories. Measured 2026-09-13 by a script that walks every local git checkout on the machine and counts unique commit hashes by author over the trailing year. Of that, about 31,000 commits sit under my name across all 58 repos, and about 9,200 sit under my brother Jacob’s name across 7 repos, 8,750 of those inside one repo, Connections. One day, 2026-09-05, hit 966 commits by itself. Commit activity is flat around the clock, not clustered in working hours. Be clear about what that means: those are not 42,400 commits I typed. They are the output of AI coding agents I run and direct, at a scale and a schedule no person works. That is not a caveat. That is the whole point of this program.

Connections, the flagship codebase

One monorepo, Connections, carries 26,011 commits in the last 6 months, 25,029 tracked files, 34 separate service planes, 443 SQL migrations, 4,924 test spec files, 1,258 automated guardrail checks, 60 automation hooks, and 26 written rule files. Read directly from the repository on 2026-09-13: commit log count, tracked file count, and directory counts for the planes, migrations, specs, checks, hooks and rules. It has not launched publicly yet. This is the production codebase the demo and the curriculum are both pulled from.

Tokens and API-list cost

245.8 billion tokens processed through this one machine between February and September 2026, $138,994 at API list price. Measured 2026-09-13 with ccusage run over the local session transcripts on this machine. That dollar figure is an undercount on two counts: it is one machine out of several that run agents, and at least one model used does not have a public list price to apply, so its tokens are not priced at all. The real number, across every machine and every model, is higher than what is shown here.

Doctrine, in words

About 1.1 million words of written, dated operating doctrine: 235,000 words in a shared global memory store of 424 files, 406,000 words spread across per-repo memory files, 470,000 words inside Connections’ own memory, 51,000 words of standing rules, 10,000 words of role guides, plus a 6,500-word global instruction file that every agent reads first. Alongside that sit 74 purpose-built tools, 59 skills and 60 automation hooks. Measured 2026-09-13 with a word count over every file in the memory and rules directories of the shared memory repository and the Connections repository. This is not a wiki nobody opens. It is what the agents read before they act, and it gets rewritten every time something breaks.

AgentHydra

179,000 lines of code across 1,096 files, a shipped, downloadable desktop product with 43 stars on GitHub. Counted 2026-09-13 from the AgentHydra source tree directly, with the star count read live off its GitHub repository page. This is the fleet console in the demo: the dashboard that shows every local Claude, Codex and OpenCode session in one place.

SageThumbs 2K

14,096 release downloads on GitHub, 176 stars, first released June 2026, signed by LUNARWERX LLC. Read live off the GitHub releases page on 2026-09-13, straight from GitHub’s own download counter and star count. A handful of other public repos carry real stars too: FastColabCopy 33, DevWebUI 31, RepoYeti 29, AnatomyOf 29, NormWind 27, ReDesign 23, YTSort 13, same source, same day.

The GitHub graph gap, and why it exists

Our public GitHub profiles show 886 contributions for 2026 between us, against 26,011 commits in Connections alone over the same period, a figure GitHub’s own API already attributes to my account. Read live off both GitHub profile pages and cross-checked against the GitHub API on 2026-09-13. The gap is not fake activity hiding as real activity. It is the opposite: real activity that GitHub is not showing, because neither of us has joined the organization the work is filed under or turned on private contribution counts. That is a two-click fix, and it is getting fixed before this page goes live, so what you see on our profiles by the time you read this should already match what is written here.

Rerun it yourself

Two things will sit beside this page so nobody has to take our word for any of it: the commit census script that produced the commit and repo counts above, and the exact ccusage command that produced the token and cost totals. Point either one at your own machine and you get your own numbers, not ours.