Lessons learned

Insights from a decade of building privacy-first solutions in 3 stacks.

I’m Alex, a software engineer and a certified tax advisor in the Czech Republic. Since 2015, I’ve worked on legacy codebases and new apps alike: C++/Qt desktop apps with FUSE file systems on macOS and Windows, iOS and Android apps in Xamarin and Swift, a linked data index, web crawlers, a tax engine, accounting integrations, and a self-hosted homelab. These days I lead two products, ship new apps to production, and coach engineers on working with coding agents.

From 2018 to 2021, I was also an accredited assistant in the Czech Parliament, mostly on digital regulation in the EU committee. I studied computer science, data science, and corporate finance with a minor in taxation. That is why the posts keep coming back to encryption, authentication, and data that carries legal weight.

Most posts start with something that broke in an unexpected way. Others share research on a topic before the failure. Either way, they are written for a person or an agent to reuse. Below is the work behind them: a year of commits and the tokens the coding agents spent on them.

Contributions as of November 8, 2026

Model usage

Combined recorded tokens across models. Snapshot: 2026-11-07 (UTC); updated weekly from imported sessions. Windows include the snapshot day.

7 days · 523,127,128 tokens
Token breakdown by model, 2026-11-01 through 2026-11-07 (UTC)
Model / shareInputCachedOutputTotal
claude-opus-5-5 · 91.1% 6,352474,469,7802,104,074476,580,206
gpt-daybreak-blue-latest · 5.7% 1,402,48928,094,336166,88929,663,714
kimi-k3 · 3.2% 377,26616,283,90489,33016,750,500
30 days · 4,160,930,898 tokens
Token breakdown by model, 2026-10-09 through 2026-11-07 (UTC)
Model / shareInputCachedOutputTotal
claude-opus-5-5 · 84.8% 44,0613,515,113,60514,269,3103,529,426,976
gpt-6.1-sol · 4.4% 6,398,274175,468,544824,605182,691,423
glm-5.3 · 4.0% 2,081,305163,664,8961,225,519166,971,720
kimi-k3 · 2.9% 2,730,301116,222,080395,743119,348,124
gpt-daybreak-blue-latest · 2.7% 6,053,719105,334,528623,356112,011,603

Only models with at least 1% of a window's tokens are listed; the others still count towards its total. Input excludes cached tokens; cached includes reads and writes. Total = input + cached + output; reasoning is included in output, not added again. These are token volumes, not costs or equal amounts of work. Imported history may be incomplete. Hermes session/model aggregates are assigned to their last recorded activity date, so window boundaries are approximate for sessions spanning multiple days. No account, provider, pricing, or conversation details are published.