I build infrastructure for people who run AI agents.

I run Kyma API, a gateway that gives developers one key for a hundred models, and Mandeck, a macOS terminal for driving several coding agents at once. Most of what I build is open source, and almost all of it exists because I needed it first and could not find it.

Most of the building is done by agents now. Four manager agents each own one project and start their own workers, and Mandeck is the room they run in. The split is by measured strength rather than preference: Claude writes the specs and the reviews, Codex and Cursor do the construction, because Claude is the scarcest budget I have. Overnight on 11 September that produced ten merged and verified changes on a single product.

I live in the Bay Area and the community I run is in Vietnam, so most of what I build has to work in both places. I ship in public with the numbers attached, including the ones that did not work out.

  • Kyma API Live

    Gives agent builders one key for a hundred models, with uptime measured rather than claimed.

    An internal audit on 2026-09-08 found one model advertising 99.6 percent uptime against 74.0 percent measured, and 32 of 70 published prices wrong. Both were corrected from served request logs

  • Mandeck Live

    Lets one developer run many coding agents at once and hand work between them.

    Launches 25 agent CLIs (Claude Code, Codex, Cursor, Gemini, Grok, OpenClaw and 19 more) plus a plain shell

  • Walks Vietnamese builders from a real problem to a product people pay for, in four weeks.

    140 enrolled, 120 signed into Campus, 61 submitted at least one assignment as of 2026-09-12

  • Affitor Live

    Runs affiliate programs for SaaS brands, from click tracking through commission holds to partner payouts.

    Live at api.affitor.com and serving the brands on it; last production deploy 2026-08-23 (v3.16.3)

All projects