Private AI · Manufacturing · Retail · Food and beverage
Itqan builds AI agents for operations, supply chain and customer teams, and runs them inside your own cloud account. The data stays in your region. A person on your team approves anything an agent sends.
Industries
For each industry there is a first agent we can pilot in three weeks. These are the jobs we look at first.
Supplier email, quality logs and shift reports.
Good first agent: supplier email · 3-week pilot
Store questions, customer messages and product data.
Good first agent: store operations · 3-week pilot
Supplier orders, recipes and catering requests.
Good first agent: catering requests · 3-week pilot
One agent, one run
Pick an industry. Each run ends the same way: a person decides.
What we build
Agents that read the inbox, draft the order, answer the store and file the report. A person approves what goes out.
Procedures, specs, contracts and past orders, answered in plain language with a link to the page.
Open-weight models tuned on your records and hosted in your region, for when a general model is not good enough or not allowed.
Weekly test runs, model upgrades, a monthly cost ceiling and a written note on the fifth of each month.
Security
Your security team sees the architecture, the data flow and the controls before we index a single document. Nothing goes live until they sign off.
Nothing leaves your account unless a person approves it.
We deploy in your AWS, Azure or Google Cloud account. You hold the keys, the code and the logs.
Processing in the UAE or the EU when the model allows it. When it does not, we tell you in writing before we start.
We only use providers and settings that keep your content out of model training.
Single sign-on and roles. A store manager sees store documents, not payroll.
Every agent action, source and approval is logged, and you can export it.
Anything that leaves the company goes past a person first.
We test every agent for prompt injection and data leaks before go-live, then rerun the tests every week.
When a customer talks to an agent, the agent says so. Article 50 of the EU AI Act requires this from 2 August 2026.
For your security review we send the architecture, the data-flow diagram, a data processing agreement and our list of controls.
Request the security packHow we work
Thirty minutes with your leadership. We ask where the hours go and what AI should never touch.
30 minutes · freeWe pick the three jobs an agent should take first, and write down the hours each returns, the data each needs and the risks.
1 week · $750One agent on one workflow in your cloud, with a test set and a person approving every output.
3 weeks · fixed price, set in the mapMore agents, private models where they pay off, and a written note on the fifth of each month.
Monthly · 30 days noticeWork
Before AI we built software, and the same release discipline carries over. Our AI case studies go up as each client signs them off.
Food producer · Gothenburg, Sweden
Millow ferments Swedish oats and mycelium into protein. We are the design partner on their two-language website, working beside their in-house engineer and releasing every week.
Onboarding and training software · Sweden
Nexilink gives Swedish municipalities digital onboarding and training on real situations for new staff in welfare services. We helped build the product, from the brand and the research to the web app and dashboard design.
“They didn’t just redesign the product. They rebuilt it into a clean, functional and user-focused experience that perfectly matched our direction.”
Our own studio runs on five agents: inbox triage every morning, a sales call turned into a proposal draft, compliance reminders, an end-of-day close and a weekly review.
Questions
Read access to the systems the first agent works with, through an account your IT team creates and can switch off. Write access only where a person approves each action.
You do. The code, the prompts, the test sets and the cloud account are yours from day one. If you stop working with us, we hand everything over within 30 days, with no exit fee.
The one that passes your test set at the lowest cost in the right region. That can be OpenAI, Anthropic, Mistral or an open-weight model on your own servers. When a new model comes out, we rerun the tests.
The AI Map is $750 and takes a week. The pilot is a fixed price that we write into the map, so you see it before you commit. Running costs are a monthly fee plus your cloud and model bills, passed on at cost.
Arabic, English and Swedish, for the agents and for the people you talk to at Itqan.
The briefing can be this week. After the AI Map is paid, kickoff is within five working days.
Company
A demo is easy. What we get paid for is the agent that still answers your suppliers correctly in month twelve, after the price list has changed twice and the model has been upgraded once. That is the one we build.
For the people who run operations, supply chain, IT and customer service in manufacturing, retail and food and beverage.
Pick a timeOr write to ibrahim@itqanstudio.com