Use cases
Five sectors, one problem in common — AI has to stay in-country and under control.
Grouped by the door each one comes through — sell what you build, or prove nothing left.
THE MAP
One wall, five sectors
Every sector on this page is stuck on the same wall: the AI they need can't sit where their data and rules require it to.
Sentinel moves that wall — AI that stays in-country and under control. What differs from one sector to the next is the promise they are really buying, so the sectors below are grouped by the door they come through.
FOR TELCOS & BUILDERS
For telcos & builders
Selling what you build on it. What these buyers want to see is a number — what it cost, what it sold for, and the margin in between. See the telco door.
Telcos & operators
The problem: your customers are already buying AI — from foreign clouds, on foreign cards, billed in dollars. You carry the network the traffic rides on and capture none of the revenue, and you have no model lab or billing system to change that.
What Sentinel does: launches an AI revenue line under your own brand. Resell Sentinel on your own markup and prepaid credits, see the margin on every request, and give each customer its own white-label look and invoice — with no model lab to run and no billing system to build. Your team works in a shared workspace with clear roles, not a shared login.
The outcome: a new revenue line you own end to end, with the margin visible on every call.
Banking & finance
The problem: your most useful data is exactly the data you can’t send to a foreign cloud, and without spend controls and redaction, AI is a risk no compliance officer will sign off. So it stays interesting but unapprovable.
What Sentinel does: runs KYC, support, treasury, and analysis on regulated data — in-country, billed in PKR, with spending caps per key, quotas per team, and guardrails that strip out personal data before any request leaves the pipeline. Spend is tracked per desk and capped per key.
The outcome: that combination is what turns AI from possible into approvable.
Developers & startups
The problem: building on AI today means a foreign cloud account, a foreign card, and dollar billing before you’ve shipped a thing — and if you grow into serving your own customers, you start the plumbing over.
What Sentinel does: gives you an API that works with your existing OpenAI, Anthropic, and Gemini tools — point your current SDK at our endpoint, pay in PKR from prepaid credits, and ship on our in-country model without a foreign cloud account. In runnable code the model is named sentinel-mesh; the console lists the exact model behind it.
The outcome: start with a single key; the same console grows to handle your own customers when you have them.
WHAT IT COSTS
What these cost, and when a lane is worth it
Which class a workload lands in depends on the shape of its requests, not the plan you bought. The in-country hardware has a fixed amount of working memory: a short back-and-forth question uses a little, a large agent request (around 30,000 words of context) uses much more — so one Context Lane takes up about the same room as eight Chat Lanes.
The bill follows the same fact. Short prompts are priced at $0.05 / 1M words of input; larger prompts (above roughly 2,000 words) at $0.40 / 1M — the same 8× step, for the same reason — and output is $2.00 / 1M. Lane prices are on the pricing page; every lane includes an allowance, and anything past it is billed at exactly these rates.
Chat class, pay-as-you-go — a subscriber support assistant.
$107a month
200,000 requests a month, each roughly 700 words in and 250 out. Every prompt is short, so input works out to $7 and output to $100 — about $107 a month. Nothing is reserved: it runs in the shared pool and gives way to reserved and defence traffic at peak times. What it costs you tracks what you use, and nothing more.
Chat class, reserved — a bank's internal desk.
$396a month
The same workload, but it must answer during the busiest hour of every day. Four requests at once at peak means four Chat Lanes: $396 a month, with the rest of the volume covered by the lanes’ allowance. You are buying the peak, not the volume — pay-as-you-go, the same traffic costs less but holds no place in the queue. Whether the reservation is worth it comes down to what a delayed answer costs you, and that is exactly the right question to ask.
Context class, pay-as-you-go — an agent used now and then.
$30a month
Large prompts, 2,000 turns in the month. Input comes to $24 and output to $6 — $30 a month — and it waits behind reserved traffic when the hardware is busy. Long, heavy requests are genuinely affordable by usage; they simply cost a bit more because they take up more room.
Context class, reserved — a coding-agent team.
$1,398a month
Two agents working at the same time through a normal workday, 8 hours across 22 days, is roughly 14,000 of those heavy turns. Pay-as-you-go that comes to $168 of input plus $42 of output — $210 a month — against two Context Lanes at $1,398. The gap is the reservation: agent work that gets stuck waiting at the wrong moment stalls a person, and two lanes are two agents that never wait. Run those lanes flat out around the clock and the pay-as-you-go equivalent climbs to roughly the lane price — a sign that both price lists were built from the same real-world measurement, not set apart.
The rule of thumb: pay-as-you-go charges for what you used; a lane buys you a place you never have to wait for. Buy lanes at the point where waiting costs you more than the reservation.
RUN IT YOURSELF
Run it on your own infrastructure
The problem: some organizations can’t send their traffic through a shared cloud, yet don’t need a sealed enclave either — and their internal teams have each opened their own API accounts, so there’s no central control, no shared view of spend, and no one place to set the rules.
What Sentinel does: that is Mesh on-premise — the same Sentinel Mesh engine, self-hosted for one organization’s internal routing.
Enterprises running AI internally
Deploy the full gateway — routing, virtual keys, spending caps, guardrails, and Sentinel Watch — on your own servers or cloud account, and point every internal team at one governed endpoint instead of a scatter of separate API accounts. You choose which providers it calls and where they run; outbound traffic is your decision, not ours. It comes as one flat offline licence rather than a meter, so the bill does not move with the traffic — and usage and Sentinel Watch still show in full, because a licence changes how you pay, not what you can see.
The outcome: one governed endpoint for the whole organization, at a fixed cost you can budget.
FOR DEFENCE & GOVERNMENT
For defence & government
Proving nothing left. What these buyers want is a guarantee they can see — built into the system, not just written into a contract. See the defence door.
Government & public sector
The problem: citizen data and internal government work can’t leave the country, and “trust us, it’s compliant” doesn’t survive an audit. Residency promised on paper isn’t residency.
What Sentinel does: runs citizen services, executive dashboards, and internal tools on AI that stays inside the country. Residency is built into the system rather than promised on paper, and Sentinel Watch records where every request ran.
The outcome: an audit trail an assessor can read, not a clause you have to defend.
Defence & national security
The problem: for classified work, any outbound connection is a risk you cannot accept — and even the metadata of usage, how much and how often, is intelligence that must not leak.
What Sentinel does: the Sentinel Sovereign enclave runs fully isolated — no outbound traffic allowed, so nothing can ever leave your network. It runs on dedicated in-country hardware carrying one customer’s traffic and nothing else: there is no shared queue and no neighbour whose activity could hint at yours. Classified traffic is deliberately not metered, because even counting it would reveal how much you use, and when.
The outcome: a sealed system where the absence of any outbound path is the proof, and nothing — not even a usage count — leaves.
Find your door.
Sell what you build under your own brand, or prove that nothing ever left — either way, it's the same Sentinel Mesh engine underneath.