AI that ships, sold by the hour and the token.
We build AI-powered websites, apps, agents and automations — and we specialise in Claude. You buy delivery capacity, not a vague retainer: a fixed number of engineering hours plus a token quota you can spend on any AI work you ask us for.
Eight things UAE companies actually ask us for
Not a menu of buzzwords. This is the work that comes through the door, ranked by how often we are asked for it.
AI-built websites and apps
Full products, from spec to production. AI compresses the build; it does not replace the review, the testing, or the person answering for the result.
Custom AI agents
Agents that use real tools — your CRM, your database, your file store — over multi-step workflows, with approval gates on anything irreversible.
Assistants over your documents
Retrieval-backed assistants that answer from your contracts, policies and manuals — and cite the page they took the answer from.
Documents into structured data
Invoices, LPOs, trade licences, Emirates IDs, delivery notes — read, validated, and pushed into the system that needed them.
Arabic-first conversational AI
WhatsApp, web and phone assistants that hold a conversation in Gulf Arabic and switch to English mid-sentence, the way your customers actually speak.
Wiring AI into what you run
ERPNext, Odoo, WHMCS, WordPress, Microsoft 365, custom back-ends. The value is usually in the connection, not the model.
Evaluation and guardrails
A scored test set before anything goes live, so “it seems to work” becomes a number you can hold us to.
Hosting and data residency
We run our own servers. If your data has to stay in a specific jurisdiction, say so at the scoping call and we design around it.
We build on Claude, and we run our own business on it
Plenty of firms will name three model providers and pick whichever is cheapest that week. We went deep on one, because depth is what makes an agent reliable enough to leave running.
Agent tooling, not just chat
Tool use, the Model Context Protocol, structured outputs and long-running agent sessions. This is the layer that turns a chatbot into something that does the job.
A one-million-token window
A whole codebase, a full tender document, three years of correspondence — in one pass, without chopping it into fragments that lose the thread.
This site is the reference
its.ae is bilingual, dark-mode, agent-maintained, and wired to our billing portal through our own tooling. We sell the method we already use.
Two meters, one invoice
Engineering hours
Our people: scoping, prompt and context design, building, integration, testing, review, deployment and handover. Billed in 30-minute increments against the work item you approved.
Token quota
What the models consume while doing your work — during our development runs and during your system's live use. One quota covers both, so you are never surprised by a second bill from somewhere else.
We are paid to keep the quota low
Prompt caching, routing simple steps to a smaller model, and batch processing for anything that is not urgent. Because the quota is bundled, saving your tokens is our interest too.
Buy capacity, not promises
All prices in US dollars, excluding UAE VAT. Every package includes the scoping call, the evaluation set, source code and full handover documentation.
Pilot
One workflow, proven or disproven, before you commit anything larger.
- One workflow scoped and built
- Measured against your current baseline
- Written go / no-go recommendation
- Reply within 2 business days
Build
A real system in production — one department, one clear job.
- Production deployment, ours or yours
- Integration with one existing system
- Scored evaluation set you keep
- Two training sessions for your team
- Reply within 1 business day
Scale
Several workflows, live traffic, and someone watching the numbers.
- Up to four workflows in production
- Monitoring dashboard and cost report
- Monthly tuning against live data
- Arabic and English coverage
- Same-day reply, office hours
Embedded
A named engineer inside your team, month after month.
- Named engineer, your stand-ups
- 25% of unused capacity rolls over
- Private or on-premise deployment
- Quarterly roadmap review
- Priority queue, 24/7 escalation
Top-ups and rates
| Item | Pilot | Build | Scale | Embedded |
|---|---|---|---|---|
| Extra block of 10 hours | 760 | 670 | 600 | 560 |
| Extra block of 10M tokens | 250 | 220 | 180 | 150 |
| Rush delivery (48 hours) | +35% | +35% | +25% | — |
| Hourly, without a package | USD 95 / hour | |||
Private or on-premise model deployment, and any work requiring a dedicated GPU host, is quoted separately. Unused tokens do not roll over on one-off packages; unused hours can be extended once by 30 days on request.
What a million tokens actually buys
“Tokens” means nothing until you can price a job in them. These are real figures from work we have delivered — treat them as ranges, not quotes.
| A job of this size | Typical tokens | Fits inside |
|---|---|---|
| Reading a 100-page tender and answering questions on it | 0.1 – 0.3M | Pilot, many times over |
| Extracting fields from 1,000 supplier invoices | 2 – 3M | Pilot |
| A support assistant handling 1,000 customer conversations | 4 – 6M | Build |
| One agent run across a medium codebase: read, plan, edit, test | 0.5 – 1.5M | Build ≈ 20 runs |
| Translating and localising a 50-page bilingual website | 1 – 2M | Pilot |
| An internal knowledge assistant used daily by 40 staff, for a month | 15 – 25M | Scale |
| A WhatsApp sales assistant, 300 conversations a day, for a month | 35 – 50M | Embedded |
Arabic text uses roughly twice the tokens of the same content in English. We account for that in every quote — it is a known cost of working bilingually, not a surprise we pass on to you later.
Four steps, in this order
Scoping call
Forty-five minutes, free. We map the workflow and tell you plainly if AI is the wrong tool for it — that answer costs you nothing.
Pilot
One workflow, built and measured against how you do it today. You get a number, not an impression.
Build
The production system, integrated with what you already run, with the evaluation set in place before launch day.
Run
Monitoring, monthly tuning, and a cost report you can read. Models change; the system stays correct because someone checks.
The ones people ask before signing
What is a token, in one sentence?
A token is roughly three-quarters of an English word — the unit the AI model reads and writes in, and therefore the unit it is billed in. One million tokens is about 750,000 English words, or around half that in Arabic.
What happens if I run out of hours or tokens?
We tell you at 75% and again at 90%, before either meter runs dry — never after. You then add a top-up block or we pause. Nothing is consumed beyond your package without your written approval.
Who owns the code, the prompts and the data?
You do — all three, on final payment, including the prompts and the evaluation set. Your data is never used to train anyone's model. If you leave us, you leave with a system you can hand to another team.
Can the models run on our own servers?
Yes, for open-weight models, and we host it either in our UAE facility or in yours. Frontier models such as Claude run as a hosted API — if a regulator requires everything to stay inside your walls, we say so at the scoping call rather than after you have paid.
Do you work in Arabic, properly?
Yes. Interfaces, prompts, evaluation sets and documentation, in Modern Standard and Gulf Arabic. Everything we ship exists in both languages from the first day, including this page.
We already tried an AI tool and it disappointed us.
Usually because nobody defined what “working” meant. That is what the evaluation set is for: an agreed test with a score, run before launch and re-run every month. If the score drops, we know before your customers do.
Bring us the workflow that annoys you most
Forty-five minutes, no charge, no slide deck. You describe how the work is done today; we tell you what AI would change, what it would cost in hours and tokens, and where it would not help at all.