I build AI products for people and businesses in the Gulf and beyond, bilingual by design: English and Arabic as equal first-class citizens, with Gulf and Kuwaiti dialect handling, right-to-left interfaces, Arabic OCR and Arabic voice. I care most about the parts that decide whether AI can be trusted in daily use: approval before anything risky runs, memory you can inspect and correct, and systems that verify outcomes instead of reporting activity.

I prefer measurement to claims. Kazma publishes its prompt-injection results on a public benchmark — including the results that don't flatter it — and keeps a dated list of its own known weaknesses.

Open source (MIT) · v0.11 · Beta until 1.0

A self-hosted AI agent platform, bilingual by design: English and Arabic as equal first-class citizens, not Arabic as a nice-to-have. One agent serves you across the web, the terminal, the CLI and your messaging apps (Telegram, Discord, Slack), and around it sits a full workbench:

• Web IDE for working in your repositories;
• X Studio for composing, scheduling and managing X posts through the official API;
• Knowledge Base that grounds answers in your own documents and sources;
• Deep Research, a multi-source pipeline that produces full research papers, not quick search answers.

It is built for developers, operators and technical teams who want an agent that actually works for them: long-term memory, human approval before anything risky runs, and your own infrastructure. It is the most mature of my projects, with more than 12,000 tests, a clean security audit, and a live install I depend on every day. I keep the public release labelled Beta until 1.0, out of discipline more than doubt.

Get started: Install · Documentation · Live demo · Security results

Python · LangGraph · FastAPI · PostgreSQL · SQLite · Textual · Alpine.js

Private

A one-place AI merchant platform, unified across channels: the merchant's entire business runs through one AI layer over WhatsApp, Instagram and the web (email included), with a single omnichannel inbox and one catalog synced everywhere, not three disconnected apps. Catalog, orders, payments, returns and customer conversations, in text or Arabic voice notes, all in one place.

• Delivery that confirms itself. It doesn't just dispatch: it confirms the courier accepted and picked up the order, and when a driver refuses or is busy it offers the run to the next available courier, with no one chasing drivers by hand.
• Real-time fraud screening. It watches order patterns (velocity spikes, unusual-hour bursts, abusive ordering) and sends high-risk orders to a human instead of dispatching them.

Starting in Kuwait, then the GCC and the wider Middle East, on an architecture kept deliberately global-ready.

Private

"All-in-one" is the whole point. It serves trainees, and coaches run their rosters on it; supplement stores and diet restaurants sell into its marketplace, and gyms get their own tenancy. One ecosystem built around the training plan.

Private · before Beta

The biggest and most ambitious of my projects: an AI operating system for companies. It doesn't just hold tasks — it follows them. Every employee gets a workspace and a sub-agent that turns conversations into tasks, tracks progress and nudges with reminders. The system watches for stalled work, bottlenecks and overdue items, guides people through how to accomplish what is assigned, and verifies "done" by evidence: activity is not achievement. Admins get their own management layer and roles, and the founders see verified completions and escalated exceptions instead of noise. Around that core: multi-agent feasibility analysis for new ventures, continuous risk monitoring with bilingual alerts, and company knowledge graphs. It needs the most time before Beta, and I'd rather give it that time than rush it.

Open source

A portable, ultra-secure personal vault for secrets, API keys and private knowledge, with a local (Ollama) or cloud AI assistant. TypeScript.

Pilots, partnerships and collaboration: admin@kazma.ai. On X: @KazmaAI for Kazma, @b_alfaris personally. Security reports for Kazma: private advisory.