For telcos & builders
Sell it as your own.
Your subscribers want AI, and you want it on your own bill — a new revenue line under your brand, not a foreign cloud you hand your customers to.
THREE THINGS YOU DON'T HAVE
Sentinel removes all three
You buy sovereign AI capacity at a wholesale price and set your own retail price.
But standing that up means three things you don’t have: a model lab to run, a billing system that can rate and invoice every request, and a way to prove your margin customer by customer. Most operators stall right there.
A model lab to run
No model lab to run.
A billing system that can rate and invoice every request
No billing stack to build.
A way to prove your margin customer by customer
Every request is priced, metered in PKR, and itemised — what it cost you, what you charged, and the margin between the two — for each customer, each month.
No model lab to run. No billing stack to build. Your brand on everything your subscribers see.
ONE GATEWAY, TWO DEPTHS
Start as a developer. Grow into an operator.
The same gateway serves both, so nothing has to be rebuilt when you scale up.
Day one — one key. Point an existing OpenAI, Anthropic or Gemini SDK at sentinel.junaid.pk/v1, pay from prepaid PKR credits, and ship against our in-country model without a foreign cloud account. See the Quickstart.
Later — customers. The same console grows an admin side: set each customer’s markup, credits, limits and keys — each one kept separate, each one invoiced. See Reseller.
MARGIN, PROVED PER REQUEST
What you buy: wholesale capacity
You buy capacity — the number of requests the sovereign cloud handles at the same time. It is real, finite inventory you can count and plan against, the way you already plan spectrum or trunk lines. You buy it wholesale and resell it at your own price.
Capacity comes in two sizes, and the difference is measured, not guessed. A short chat request is light. A long agent request — one that reads tens of thousands of words at once — is heavy, and uses about eight times the capacity. So it costs about eight times as much.
| What you buy | Carries | Weight | List / month |
|---|---|---|---|
| Chat Lane | Short conversational and API traffic | 1 | $99 |
| Context Lane | Long-context and agent workloads | 8 | $699 |
Each unit includes a monthly usage allowance; anything beyond it is billed at the published rates. Your wholesale rate sits under this list price and is set in your agreement. Your retail price is yours to set on top, per customer, in PKR.
Because heavy capacity is priced for exactly what it uses, your margin is the same shape whichever size you sell. You are not guessing at a premium for long requests, and when a customer shifts from simple assistants to heavier agents, you can see the change and re-price it.
Two points worth having in a commercial conversation:
- What you sell is capacity that exists. The available pool is measured, and we sell only up to two-thirds of it. When it is full, new orders are refused rather than accepted and quietly queued.
- A request is billed for what it uses, not the plan it was sold under. A customer on a chat plan who starts sending very long requests is billed for the heavy work it really is — so your cost and their bill never drift apart.
Pay-as-you-go traffic, and any usage past an allowance, is billed on the same rate sheet on the pricing page: $0.05 / 1M input for short prompts, $0.40 / 1M for long ones, $2.00 / 1M output.
That step up in price for long prompts is the same 8× that prices heavy capacity, for the same reason: a long prompt costs eight times as much because it takes up eight times the room — whether you buy it as capacity or by the request. One explanation covers your whole rate card. A customer who questions the long-prompt rate gets a straight answer about hardware, and you are never defending a blended price that quietly lets your lightest customers subsidise your heaviest.
THE TELCO CASE
What you are actually buying
A price you control
wholesale × your markup × FX, applied as the bill is priced rather than reconciled afterwards.
Margin you can see on every request
not a monthly total you have to take on trust.
Inventory you can plan against
capacity is a fixed, published pool, so how much you can sell is a number, not a hope.
A white-label surface
your logo and colours across the Playground and Platform your subscribers use.
Separate customer accounts
an operator console at /app/admin for markup, credits, limits and keys.
Limits that hold
every key rate-capped, every customer bounded by a quota. AI spend stops being the line item nobody can put a ceiling on.
WHO'S BUILDING THIS
Built by people who already build telco systems
Sentinel is not an AI company’s first encounter with a mobile network. Mercurial Minds has a telecom practice that predates it — the unglamorous core, not just the demo layer:
| Core network | OSS / BSS and service layer · SS7 · SIGTRAN signalling |
| Messaging & VAS | SMSC · iIVR · digital VAS · 5G solutions |
| Subscriber-facing | VoIP and self-care apps · call-centre and GSM location solutions |
| Integration | WSO2 API and identity · CRM · Samsung KNOX · streaming |
| Operations | Managed 24/7 NOC monitoring · operations and maintenance |
12+ years, 18 countries, 300+ projects across telco, banking and fintech, energy and government — and a collaborating firm of Andersen Consulting, so global advisory reach with delivery that stays local.
Why that matters more than it sounds
Reselling AI is not an API problem. The API is the easy part.
The hard part is everything the API has to survive contact with: identity (who is this subscriber, and on which plan), billing (rate it, invoice it, handle the dispute), provisioning (switch it on for a group of customers without a migration), support (a front-line agent has to explain a charge), and the NOC (something pages an engineer at 3am, and that engineer needs a runbook).
Those are the systems Mercurial Minds already builds. When Sentinel meters a request, prices it against a customer’s markup, and puts it on an invoice line, that design comes from people who have been on the operator’s side of the integration — not from an inference vendor guessing what a network needs.
What you can ask for beyond the gateway
Because the same firm builds both, the integration work is not somebody else’s problem:
- Wiring the gateway into your existing identity and billing systems, rather than running a parallel billing island.
- Self-care and CRM surfaces so subscribers see AI as part of your product, not a separate portal.
- Contact-centre automation on the same in-country AI — one of M.M’s five practice areas, alongside IT, HR, Finance and Cyber Security.
- NOC-grade operations — the monitoring discipline that already runs 24/7 networks, applied to the AI you are reselling.
GOVERNANCE
Your team, scoped
Running a fleet is a team job, so the console is multi-user. Your people share a workspace with real roles — a manager runs members, spend and keys; a developer builds against the gateway and can read the balance — and the workspace someone is working in decides whose credits a request spends. The person who created the workspace holds an owner protection: they cannot be demoted or removed, so your account can never be locked out of its own administration.
The reseller operator role is scoped to your fleet only: your customers, markup, credits, limits, capacity allocation, margin. It deliberately cannot reach the routing engine or another reseller’s traffic.
That scope is enforced, not just shown on screen. A single access policy decides every action from who is asking, what they are trying to do, and which customer it touches — and list views are filtered to your fleet, so the customers, credits and margin you can see are exactly your own and never another reseller’s. Finer control follows the same rule: a developer can revoke only the keys they created, so one engineer can never cut off another’s live traffic by accident. See the whole model under Sentinel Platform.
THE BOTTOM LINE
Why this is a revenue line, not a cost centre
AI spend is usually invisible until the invoice lands, and impossible to trace back to the customer who caused it. Sentinel turns that around: the same metering that prices your cost prices your customer’s bill, so margin is built into every request instead of arriving as a quarterly surprise.
Underneath is the Sentinel Mesh gateway — one engine, offered three ways. This door is that engine run as a metered service on our in-country cloud: you resell it under your own brand, a business builds on it directly, or a consumer tries it free in the Playground. The same gateway also runs self-hosted on a customer’s own servers under one flat annual licence, for an organisation that wants it inside its own walls, and as the defence build behind For defence. What changes is where it runs and how it is billed — not what is running.
Sell it as your own.
AI spend is usually invisible until the invoice lands, and impossible to trace back to the customer who caused it.