Chams Bouzaiene
software engineer · hamburg, germany
I build AI agents. Most of what I ship is some version of the same question: can software reach into the messy systems people actually use — browsers, files, chats — and get real work done without taking your data hostage?
Right now I'm focused on BinaryClerk, a local-first desktop AI coworker. One app on your machine that chats with capable models, drives your real Chrome through a companion extension, and keeps projects, chats, and files on disk instead of in someone else's cloud.
Before that, I built a handful of smaller open-source agents: dodo (a coding agent in Go), CtxAnt (chat with a browser agent over Telegram), and QA-agent (an agent that proposes QA operations for developers). Each one taught me something I'm carrying into BinaryClerk.
I write here when I learn something worth keeping. Mostly: agent loops, LLM behavior under load, local-first patterns, and small tooling. No hot takes.
background
I'm a software engineer based in Hamburg, shipping production systems and developer tooling with a bias toward TypeScript, Go, and pragmatic architecture. I care about reliability, observability, and giving users control over their data — especially now that AI features are landing in every product surface.
what drives me
Local-first and BYOK patterns, honest security write-ups, and agents that automate boring work without hiding failure modes. I want tools that feel fast, inspectable, and owned by the person using them — not rented from a dashboard you don't control.
currently
Shipping BinaryClerk in public, tightening the waitlist and extension pairing story, and writing longer technical notes when something in the ecosystem deserves a careful teardown — like supply-chain attacks on the tools we use every day. See also /now for a snapshot of focus.