pricing

Know what your team’s AI costs — before the invoice, not after.

Capped at the machine, so the overage never happens. Signed by the machine that did the work, so finance does not have to take our word for it.

$8,400

against a $2,000 budget

What one 30-person startup’s AI invoice came to. A 4.2× overage nobody saw until it arrived.

98%

of FinOps teams now manage AI spend

Up from 31% two years ago. Somebody already owns this line, and is already being asked about it.

$200–500

per developer, per month

What a senior engineer’s AI tooling now runs to. Budgeted per person, reconciled monthly.

Most tools tell you afterwards

They read your vendors’ invoices and show you what already happened. That is a different job from the one below.

the usual

Show you what you spent, after the vendor billed you.

here

Stop it at the machine, before it is spent.

A ceiling you set, on by default at a conservative number, enforced by your own hardware — not an alert that arrives with the invoice.

the usual

A number in a dashboard you have to trust.

here

A receipt the machine signed, that you can verify yourself.

Every unit of work is signed and chained where it happened. Your finance team checks the signature; they do not check our reputation.

the usual

Report ₹0 for work that ran on your own machines.

here

Count it, at the same rate as anyone else’s.

No invoice is issued for your own GPU, so every cost tool reads it as free. The watts were real. We are the only ones who can price the whole picture.

Three, and you will know which

Named for who is reading, not for what is switched on. You start at the top and move only when something makes you.

Yours

One person, any number of their own machines.

₹0

available now

Not a trial and not a ceiling you grow into. There is no feature switched off waiting for a payment.

What your own AI work cost, computed on your machine and sent nowhere.
A spend ceiling, on from the first minute, enforced by your own hardware.
Every unit of work signed where it happened, yours to keep and to check.
The whole fabric — every model, every operator, every agent — running locally.

Run a gram →

Team

A team that has to answer to somebody for its AI spend.

₹500

in build

per developer, per month · minimum 10 · billed on an ordinary GST invoice

One number for the whole team, instead of one per person who remembers to look.
Thirteen months of it, kept, so a quarter can be compared to the last one.
A page you can send to finance, and a signature they can check without us.
Per-developer ceilings, set by you, enforced before the spend rather than after.

Take a plan →

Perimeter

An institution where nothing may cross the boundary.

Quoted

talk to us

Against the spend it accounts for, not against our cost.

Nothing settles per run, so no record of what you did leaves your network.
A deterministic bill, computed from what you licensed rather than sampled from what you ran.
Adding a person costs ₹0 — because behind a perimeter nothing about them reaches us.
A four-hour Sev-1 response, and a break-glass procedure that needs no telemetry.

Talk to us →

One developer, one repository, eight days measured on a real machine: US$807 of agent spend at list. Ten developers is over ₹2 lakh a month being accounted for. ₹500 each is about 2% of it — the same share this kind of tool charges everywhere, and a number you can verify from your own invoice.

And your own machines stay free

Your own machines are free. Forever, at any number, with or without a plan.

Not generosity, and not a promise we could quietly withdraw. The paywall is a machine boundary: if the work runs on your hardware there is nothing for us to bill, and no capability is removed from the software to be sold back to you.

What we do not sell

Every other page here makes claims. This one states limits, which are the kind of thing you can hold us to.

Uptime

It runs on your metal. We cannot restart it and cannot see it fail, so every commitment we make is a response time and the contract says so on page one.

Capacity

Nothing here is a GPU-hour rented from us. That is why there is no column on this page comparing us to a cloud.

Your own record back

Stop paying and your grams keep running and your history stays on your machines. You lose the fold and the hosting of it — not your work, and not the software.

A feature we took away

Nothing is removed from the binary in order to be sold back. If you find something gated by a plan, that is a bug and we will treat it as one.

Not true yet

The team fold is the one part of this still being built. Everything on the free tier is running today.
There is no way to put money into a gram, so nothing metered is billed to anybody yet.
Buying a plan puts no spendable balance on any machine, and is not meant to.
Nobody has been paid for serving work to somebody else. That is a different product and a later one.

A list nobody can falsify is worth nothing, so this one sits on the page rather than in a footnote.

Start with your own number.

Run a gram on a machine you already own. It will tell you what your own AI work cost, and we will never see it. Come back when that number is worth handing to somebody else.

Run a gram Verify us first

Fifty GramX Networks Pvt Ltd · prices in INR, exclusive of GST