I'm a software engineer in Hamburg, building AI agents: a coding agent in Go and a browser agent you text from Telegram. Lately I've been pointing agents at real codebases to find real vulnerabilities.
I like agents that do actual work: tight tool loops, sandboxed execution, honest measurement, and results anyone can reproduce. I post what I learn on X at @ChamsTech.
|
Open-source AI coding agent written in Go. Commands run in a Docker sandbox, it works with any LLM provider, and there's no LangChain or agent framework underneath, just its own engine. |
Text any agent on Telegram and it drives the Chrome you're already signed into. Use 12 starter agents or build your own in 30 seconds. Everything runs on your Mac. ctxant.com → |
|
An LLM audit of nginx's rewrite module rediscovers a heap buffer overflow, confirmed with an AddressSanitizer PoC. Prompts, PoC and methodology are all included, along with an honest note on what the AI did and didn't do.
|
Private for now, public when they're ready:
|
Agent loops · tool calling · MCP · browser automation · sandboxed execution · LLM-assisted code auditing
Earlier work: web, blockchain and tooling from before the agent era
| Project | What it is |
|---|---|
| react-component-lib-boilerplate | Starter kit for shipping a React component library |
| Poa-Blockchain-Docker-Compose-env | Spin up a proof-of-authority Ethereum network with Docker Compose |
| ethereum-bridge | Link private and public chains through events and a message queue |
| SVG-Editor | SVG editor in vanilla JavaScript |
| D3Export | Export D3 charts and SVGs to PDF and images |
| san-kou7.JS | A platformer game made in Tunisia, in pure vanilla JS |
| QA-agent | AI agent that suggests QA steps for developers |
| AIinfluencerNet | Platform for running a network of AI influencers |
Building agents too? My DMs on X are open.




