CTO with a passion for governed Agents
I build AI agents that ask before they act,from Dubai.
Most agent demos look great until they meet a real company: real data, real permissions, real people. I build the other kind, where the safety model is part of the design, not a slide at the end. Then I open-source it, so you can read every line and run it yourself.
| Demo | What it shows | |
|---|---|---|
 |
Zolt ⚡ | A desktop agent for macOS that talks, acts and takes meeting notes, with an Allow/Deny check before every action, secrets it can never read, and meeting text that can't trigger tools. Yes, it speaks in an 80s robot voice. |
 |
Zamil 🇸🇦 (زميل) | The same agent, Arabic first: it understands spoken Arabic, replies in Arabic, and its whole interface runs right to left. |
 |
Fitrah (فطرة): hereditary trust for autonomous AI agents. Can an agent be born with trust it can't shed, and never hold more authority than its parent may delegate? Enforced without relying on the model. |
- Ask before acting. A human approves anything that changes the world, and the approval shows exactly what will run.
- Untrusted text is data, not instructions. Web pages, files and meetings can inform an agent, but they can never command it.
- Arabic is a first-class language, not a translation added later.
- Run it in 5 minutes, or it isn't a demo.
Claude Code · Microsoft Agent Framework · Azure AI · local models · Swift · Python
LinkedIn · Dubai, UAE
Open to talks, workshops and collaborations on agentic AI in the region. Say hi on LinkedIn.
مدير تقني نهاراً، وصانع وكلاء ذكاء اصطناعي ليلاً. أبني وكلاء مفتوحي المصدر يستأذنون قبل أن يتصرّفوا، بالعربية والإنجليزية، من دبي.