01
Enquiry triage
Every form, email or WhatsApp message is read, tagged by type, language and urgency, and routed to the right person. No more shared inbox roulette.
ChatGPT integration in Dubai inside the software you already use.
You do not need a new product to use AI. We add a language model to your existing website, app, CRM or ERP for one job at a time: drafting replies, summarising records, classifying requests, writing product copy, or searching your data.
FromAED 4,950
Time2 to 4 weeks
ModelsOpenAI, Claude, Gemini
Usage costCapped and monitored
ChatGPT integration means connecting a language model, such as OpenAI's GPT models or Anthropic's Claude, to your own software through an API, so it does a specific job inside your workflow. Our integrations start at AED 4,950 and take 2 to 4 weeks. Larger ones with several features reach about AED 17,500, excluding 5% VAT.
Updated · Prices in AED, excluding 5% VAT
- Fixed price in writingScope and AED price agreed before work starts
- You own everythingStore, domain, code and accounts in your name
- 30-day fix windowBugs after launch are fixed at no charge
- Reply in one working dayFrom a senior person, not a form robot
What integration means
The chat window is the least useful part
Most people meet AI as a chat window, so they assume integrating it means adding a chat window to their site. Usually the bigger win is invisible. A model can read every incoming enquiry and tag it by type and urgency. It can summarise a 40-message customer history before your agent picks up the call. It can draft a quote email from the fields already in your CRM, in English or Arabic, for a person to check and send. Each of these is a small, well-defined feature with a clear input and output, which makes it testable, cheap to run, and easy to switch off if it does not help. We build those, one at a time, inside the screens your staff already use.
- One feature per job, with a clear input and output
- Lives inside your existing app, CRM or admin
- A person reviews anything that reaches a customer
- Can be switched off without breaking anything else
Common integrations
What UAE businesses ask us to add
02
Reply drafts
Draft answers in the customer's language from your policies, for an agent to edit and send.
03
Record summaries
A three-line summary of a customer, deal or ticket history at the top of the screen.
04
Search your own data
Ask the system in plain words, such as which customers in Sharjah have not ordered since June, instead of building a report.
05
Content at scale
Product descriptions, meta descriptions and translations generated from your data and reviewed before publishing.
06
Data extraction
Names, amounts, dates and TRNs pulled out of PDFs and emails into structured fields.
Which model
OpenAI, Claude or Gemini: it depends on the job
We are not tied to one provider. We test the candidate models on a sample of your real data and pick on quality, cost per task, speed and data terms. Many builds use a small, cheap model for routine steps and a stronger one only where needed.
| What we look at | Why it matters | |
|---|---|---|
| Answer quality | Scored on 50 to 100 of your real examples | Benchmarks on the internet are not your data |
| Arabic | Tested on your Arabic content and customers' wording | Quality varies between models and tasks |
| Cost per task | Measured in fils per request at your volume | Decides whether the feature pays for itself |
| Speed | Time to first word and full answer | Matters for live chat, less for overnight jobs |
| Data terms | Business terms on training and retention | Recorded at build time and rechecked |
| Where it runs | Provider region options, cloud platform availability | Some sectors have rules on processing location |
Model prices and names change often. We design the integration so the model can be swapped without rewriting your software.
What makes it production-ready
The parts a quick demo leaves out
A working prototype takes an afternoon. A feature you can rely on with real customers takes the rest of the work, which is what you are paying for.
Cost control
- Monthly spending cap at the provider
- Per-user and per-day limits
- Alerts at 50% and 80% of budget
- Caching of repeated requests
Quality
- Test set from your real data
- Scores recorded before and after each change
- Prompt versions kept in your code repository
Safety
- API keys on the server, never in the browser
- Input filtering against prompt injection
- Personal data stripped where the task allows
Fallbacks
- Timeouts and retries
- Second provider if the first is down
- Plain non-AI path so work never stops
Fine-tuning, RAG and prompts
Most businesses do not need to train a model
Fine-tuning means retraining a model on your examples so it learns a style or a narrow task. It is sold often and needed rarely. For most business tasks, a clear instruction, a few good examples, and the right passages from your own documents (known as retrieval, or RAG) give better results, cost less, and update the moment your data changes. We consider fine-tuning when you have hundreds of high-quality examples of a repetitive task, the output format is strict, and the cheaper approaches have been tried and measured. If someone quotes you fine-tuning as step one, ask what they tried first.
- Prompts and examples first, measured on your data
- Retrieval from your documents when facts matter
- Fine-tuning only after the cheaper options are measured
- Prompt engineering is part of every build, not a separate fee
How we work
From one feature to a measured result
- Week 1
Pick the feature and the measure
We agree one job, what good output looks like, and how we will measure it, for example minutes saved per enquiry or share of drafts sent unedited.
- Week 1 to 2
Test models on your data
We run two or three models on a sample of your real inputs and show you the scores and the cost per task.
- Week 2 to 3
Build into your system
The feature is added to your app or admin on a staging copy, with cost caps, logging and fallbacks.
- Week 3 to 4
Pilot and hand over
A small group uses it for real work. We compare the results with the measure we agreed and adjust before wider release.
Pricing
Fixed price per feature
We quote after seeing your system and a sample of the data. If your software has no API or is a hosted product we cannot change, we tell you on the first call.
One AI feature
AED 4,950from
One job, one system, measured on your data.
- Model testing on your sample
- Server-side integration
- Cost caps and alerts
- Logging and fallbacks
- Two weeks of post-launch tuning
Several features
AED 17,500typical
A set of features across one product or several systems.
- Shared prompt and test library
- Two providers with failover
- Usage dashboard by feature
- Arabic and English tuning
- Documentation for your developers
Model usage is paid directly to the provider at their rates. Care from AED 395 a month covers model updates and test reruns. Prices exclude 5% VAT.
Who we are not for
When we are the wrong choice.
A fair view, so you do not pay for more than you need.
01
ChatGPT in a browser tab is enough
If staff only need help writing emails, a ChatGPT business plan does that without any build.
02
Data you cannot share with a model provider
If your rules forbid sending data to OpenAI or Anthropic, check that first. We explain the options; your policy decides.
03
No clear job for it to do
An integration does one defined task. If you are still exploring, an AI consulting session gives you a ranked list first.
Based in Dubai, working remotely
Zelqori is a Dubai-based ecommerce and performance marketing agency.
Zelqori builds ecommerce websites and runs performance marketing for premium brands across Dubai, the rest of the UAE, the GCC and globally. We are based in Dubai and work remotely by design: projects run on Gulf Standard Time over Google Meet, WhatsApp and a shared project channel, and we meet clients in Dubai by appointment.
- Based in DubaiDubai time (GST), replies within one working day
- Remote by designGoogle Meet and WhatsApp, meetings in Dubai by appointment
- Dubai, GCC, globalUAE, Saudi Arabia, Qatar, Kuwait, Bahrain, Oman and beyond
- One teamThe store, the ads and the tracking built by the same people
A store only sells when the right people see it. See our performance marketing →
Design studies
Work that shows how we think.
Stores and systems we designed and engineered to show how we approach a brief like yours.
Questions
Answers before you call.
Prices are in AED and exclude 5% VAT.
Our integrations start at AED 4,950 for one feature in one system and take 2 to 4 weeks. Several features across a product typically reach AED 17,500. Model usage is paid to the provider at their rates, and prices exclude 5% VAT.
Yes, if we can change its code or it has an API. Shopify, WordPress, custom web apps and most CRMs and ERPs with an API can be extended. Closed hosted products sometimes cannot.
It depends on the task. We test both, and sometimes Gemini, on your real data and compare quality, Arabic handling, cost per task and data terms, then recommend one. The integration lets you switch later.
Every build has a monthly spending cap at the provider, per-user limits, budget alerts and caching for repeated requests. You see the cost per task before launch.
Rarely. Clear instructions, examples and retrieval from your documents usually work better and cost less. We consider fine-tuning only after those are measured and fall short.
Both offer business terms that state API data is not used to train their models by default. We record the terms at build time, send only what each task needs, and keep keys on the server. Confirm sector rules with your adviser.
We include it in every build, because prompts only mean something when tested on your data. If you have an existing AI feature that performs badly, we can review and fix it from AED 2,450.
Read next
Start a project
Tell us the one job you want AI to do
We will tell you whether it is worth building, which model we would test, and what it would cost per month to run.