Thinky Intelligence · Business Solutions

Thinky Intelligence — private AI for organisations that keep their own data.

Thinky Intelligence runs as one private instance per organisation — on Thinky-managed infrastructure, or inside your own network. It keeps a live, read-only copy of the records in the business systems already in use, holds an approved internal knowledge base, and answers staff questions from both, in Arabic and English, inside whatever each person is allowed to see. A retail group in Qatar runs it in production. This overview is how it is built and how it is deployed.

One private instance Arabic and English Your infrastructure or ours Delivered phase by phase

Thinky Trading and Marketing W.L.L., Doha · create@thinkycreates.com · +974 5572 6514

01The modules

Three modules, one private instance.

Thinky Intelligence ships as three modules on a single instance. Sync keeps a current copy of the data in the systems a business already runs. Core holds the approved knowledge and the reporting on top of it. Assist is where staff ask their questions. Deployments normally start with one module against one system and widen from there, because that is what the data allows — the modules do not have to arrive together.

Module	What it does	Maturity
Thinky Sync Data layer	Keeps a scheduled, read-only copy of the records in the systems already in use — ERP, finance, HR, retail and online sales — and consolidates them into one store.	Running todayScheduled sync jobs that read an ERP and an online store, keep their own copy, and report their health. Per deploymentConnectors for the systems in scope, and one consolidated store across companies.
Thinky Core Knowledge and intelligence	Turns that copy, plus an approved internal knowledge base, into dashboards, reports and answers that name the source behind every figure.	Running todayAn approved, versioned knowledge base the AI answers from; AI data chat over synced data; dashboards. Per deploymentGroup-wide reports and insights once the data is consolidated.
Thinky Assist Staff access	One place for staff to ask questions and get answers drawn from company data, scoped to their role and recorded in full.	Running todayArabic and English, roles and scopes per business unit, every answer naming its source, full audit log. Per deploymentPermissions mapped to your organisation and, where needed, to the source systems' own access rules.

Your servers

Installed in your data centre or private cloud. Nothing is shared with another company. You hold the keys, the backups and the data.

Your choice of model

A commercial model through an API, with zero-retention terms arranged with the provider, or an open-weight model running entirely on your own GPUs. Decided with you, not for you.

Bilingual from the start

Employees ask in Arabic or English and are answered in the language they wrote in, from the same knowledge.

02Architecture

How it fits together.

Your servers · your data centre or private cloud

ERPOdoo · ERPNext · SAP

Finance · HROracle · Sage · in-house

Retail · onlinePOS · Shopify

CRM · serviceAny API or database

DocumentsPolicies, manuals, files

↓

Thinky Sync · data layer

Connectors · scheduled sync jobs · one consolidated store · health check on every job

Read-only from the source systems. Each job keeps its own copy, so answers are fast and the source systems are never slowed down.

↓

Thinky Core · knowledge and intelligence

Approved knowledge base · data chat and reports · dashboards and insights

Language model: an API with zero retention, or a model on your own GPUs.

↓

Thinky Assist · staff access

Ask in Arabic or English · answers only from what the person may see · every answer explains its source

Web app on the intranet, and optionally WhatsApp for field staff · roles and scopes per company · full audit log

Everything inside this boundary runs on infrastructure you own. The only thing that may leave it is a request to a language model — and only if you choose an API model rather than one on your own hardware.

Read-only, never in the way

The platform reads from the source systems on a schedule and keeps its own copy. They are never written to and never slowed down. An answer in the chat is served from the copy, in under a second.

Every job is watched

Each sync shows when it last succeeded and when it is next due, and is flagged only when it is meaningfully overdue. You know a number is fresh before you trust it.

03Module · Thinky Sync

Sync — a current copy of the numbers, and proof that it is current.

Sync runs a scheduler that reads a business system every few minutes — invoices, stock, purchases, products, the ledger — and keeps a copy it can answer from instantly. Its health page shows every job, its interval, its last success and whether it is overdue. Where an organisation runs several systems, a connector is added for each as it is reached, and one consolidated store means a question across the whole business is a single query.

System health · jobs and sync. Each job shows its schedule, when it last succeeded, and is flagged only once it is meaningfully overdue against its own interval.

Job	Runs	Health	Last success
Sales invoices	Every 5 minutes	Healthy	2 min ago
Stock levels	Every 10 minutes	Healthy	4 min ago
Purchase data	Every 15 minutes	Healthy	9 min ago
Products and prices	Every 30 minutes	Healthy	21 min ago
ERP write-back	Every 6 hours	Healthy	3 h ago
HR records	Daily	Overdue	31 h ago

The jobs view as it runs today. The HR row shows what an overdue job looks like — it is flagged, not hidden.

System	How we connect	Status
ERPNext	API connector, scheduled jobs, write-back optional	Running today
Shopify	API connector — orders, customers, products	Running today
Odoo	XML-RPC / JSON-RPC connector, same job pattern	Built on request
Oracle	Read-only database views or REST, agreed with the DBA	Built on request
SAP	OData / RFC read access, or scheduled extracts, agreed with the SAP team	Built on request
Microsoft Dynamics · Sage	REST or read-only views, same scheduled job pattern	Built on request
Microsoft 365	SharePoint, Exchange and Teams files for the knowledge base	Built on request
Documents	Policies, manuals and files uploaded to the knowledge base — .txt, .md, .docx, .xlsx	Running today
Shared drives · PDF	Folder sync from Microsoft 365 or Google Drive, and PDF ingestion	Built on request

04Module · Thinky Core

Core — answers from approved knowledge, not from the internet.

The assistant answers from two things: a knowledge base of articles your people wrote and approved, and the synced data. It never answers from general internet knowledge about your business. Each article belongs to a company (or to the group), to a language, and carries a version, the author, the approver and the date — and only a Live article is used. A change to a policy is a new version waiting for approval, and the old wording stays in use until somebody signs the new one off.

AI knowledge · articles the assistant may answer from. Written per company and per language. Only a Live article is used.

Article	Scope	Status	Version
Leave and attendance policy	Group HR	Live	v3 · approved
Expense claims — Company A	Company A	Live	v2 · approved
Procurement thresholds	Group Finance	Pending changes	v5 · awaiting approval
IT security basics	Group IT	Live	v1 · approved
Sales commission — Company C	Company C	Draft	v1 · not in use

Draft and pending articles are invisible to the assistant — the status column alone decides what employees are told.

Knowledge base

Policies, procedures, product facts, branch and contact information — written once per company, approved by a second person, versioned forever. Bulk import from existing documents.

Data chat and reports

Ask a question of the numbers — receivables, stock, sales by branch — and get the figure with its source and freshness. Dashboards for the management layer, and data exports.

A simulator before go-live

Ask the assistant anything as a test employee and see exactly which article or table it used. Every change to knowledge is tested before a real employee sees it.

05Module · Thinky Assist

Assist — the same question, answered according to who is asking.

Every employee has a role and a scope: which companies, which kinds of data, which documents. The assistant reads only within that scope, and says so. Under every answer is an explanation panel — which data it read, which article it used, what it decided and how long it took — so a wrong answer is traceable to a wrong source in seconds, and an auditor can see why any answer was given.

Finance manager · Company A

What were Company A's receivables over 90 days at the end of last month?

QAR 1.84 million across 23 customers. Three customers make up 61% of it. The largest, at QAR 540k, has had no payment since July. Would you like the list? Source: Oracle AR ageing, synced 14 min ago · Company A only

Why it answered this

Replied in English, with one earlier message as context
Data read: AR ageing table, filtered to the asker's scope
Knowledge read: Credit control policy v4 (Group Finance)
No document outside the asker's permissions was read
Model: on-premises · 3.1 s

Warehouse lead · Company C

What were Company A's receivables over 90 days at the end of last month?

I can't help with that one. Financial figures for Company A aren't in your access. I can answer about Company C's stock, deliveries and your HR policies. Nothing outside the asker's permissions was read.

Why it answered this

Scope check ran before any data was read
Refusal names what the asker can get instead
Attempt recorded in the audit log

Employee · Arabic

كم عدد أيام الإجازة السنوية المتبقية لي؟

لديك 11 يوماً متبقياً حتى 31 ديسمبر. وفق سياسة الإجازات للمجموعة (الإصدار 3)، لا يُرحّل أكثر من 5 أيام. Source: HR system, the employee's own record · Group leave policy v3

Why it answered this

Asked in Arabic, answered in Arabic from the same knowledge
Data read: the asker's own HR record only
Knowledge read: Group leave policy v3 (Live)

Illustration of how the platform behaves: answers limited to the asker's scope, a refusal that stays inside it, Arabic answered in Arabic, and the explanation panel under every reply.

Where employees ask

A web app on your network, with two-step sign-in today and sign-on through your own directory on request. Optionally the same assistant on WhatsApp for field staff, through a number you own.

Hand-over to a person

When the assistant is not confident, or the question needs judgement, it hands over to that company's team, where a person picks it up, rather than guessing — and the employee is told so.

06Ownership and control

Your servers, your data, your audit trail.

✓

One private instance

Installed in your data centre or private cloud, or run on Thinky-managed infrastructure if you prefer to start that way. No other organisation shares the instance. Deployed on your own hardware, you hold the server access, the encryption keys and the nightly encrypted backups.

✓

Nothing leaves without your decision

Source systems are read inside your network. The only outbound call is to the language model, and only if you choose an API model — with zero-retention terms arranged with the provider. Choose an on-premises model and nothing leaves at all.

✓

Permissions that follow your organisation

Roles and scopes per company, per data type, per document set. Where a source system already decides who may see what, the platform can be made to take that rule from the source, per deployment, rather than copying it by hand.

✓

Every action recorded

Who changed which article, who approved it, who asked what and what they were told, every setting change — in an audit log that can be exported.

✓

Your data is exportable at any time

Knowledge, conversations, synced data and the log — in open formats. No lock-in by design.

The model question, answered honestly

A commercial model through an API gives the best answer quality today. A model on your own GPUs keeps every byte on your premises, at a cost in hardware and some quality. Both work with this platform, and the choice can change later. We show the same question answered both ways in the proof of concept, so the decision is made on evidence.

What we already operate

This is the same platform that runs Thinky Engage. It operates today as a private instance per customer — knowledge base, approvals, roles, sync jobs, health checks, data chat and audit log, all as described here — with a retail group in Qatar running it in production, in Arabic and English. The internal version is that platform turned inward, towards staff and the systems they work in.

07Delivery

How a deployment runs.

Every deployment follows the same shape: prove it on one unit and one system, go live there, then widen. Each stage is priced once the previous one has shown what the data actually looks like.

Proof of concept

Prove the pattern

2–3 weeks

One company · one system · 20–50 real documents · 3 roles · one dashboard · installed on a server you own.

Phase 1

First unit live

8–12 weeks

The knowledge platform live for one company with permissions, on your servers · connector to its ERP · training.

Phase 2

Breadth

Per company

More companies and systems — Odoo, Oracle and SAP connectors as each is reached · documents at scale.

Phase 3

Consolidation

After phase 2

One store across companies · group-wide insights and reports for the management layer.

The proof of concept, specifically

You choose one company and one system. The system with the simplest API is the quickest — ERPNext or Odoo, for example. You give us read-only access or a data export.

You give us twenty to fifty real documents: policies, procedures, product facts. We build the knowledge base and three roles with different permissions.

We install the platform on a server you own, connect the system, and build one dashboard from your numbers.

Your people ask it real questions, in Arabic and English, for a week. Success criteria are written down before we start.

✓

What we ask for before we start

One company and one named data owner for it. Which systems are in scope first, and how each may be accessed — API, database views or exports. Who decides permissions, and whether the model may be an API or must be on your premises. A server for the proof of concept, or permission to use one we provide inside your network.

The proof of concept is fixed-price, and its fee is credited against phase one if you proceed.

Next step: a requirements meeting.

Bring the people who own the data and the people who will ask the questions. We show the platform running live, ask the questions on this page against it, and come back with the proof-of-concept scope and price in writing.

WhatsApp or call+974 5572 6514

E-mailcreate@thinkycreates.com

Webthinkycreates.com

OfficeDoha, Qatar