Portal
Operators

Enrolment is not open yet.

Sable is building toward renting compute from machines it does not own. The control machinery now runs (serving a single node Sable operates), but the third-party half isn't open. This page exists so that when it is, we know who to write to, and so anyone weighing it up can see the hardware that will matter and the contract a node would have to speak, before spending anything.

Where this stands
Today

One gateway, one node.

Every job Sable runs executes on infrastructure Sable operates: the gateway, plus a single Sable-operated sandbox node enrolled through the node registry. That is the whole of it. No machine outside Sable's control serves any traffic.

Not open

Third-party enrolment.

The registry, the enrolment flow, heartbeats, and scheduling now run, serving Sable's own node. What hasn't shipped is everything third-party supply needs: vetting, earnings, payouts, and the compliance machinery. Enrolment stays closed until they exist.

Already published

The wire contract.

What a node would be asked to do is written down. You can read the exact shape before deciding whether you'd ever want to speak it.

Machine classes

Two classes, and only two.

Not every machine is worth putting on a network. These are the two we expect to want, chosen because each one has a buyer we can name and a guarantee we can sell. Everything else is a distraction.

ClassHardwareWhat it would serve
Confidential sandboxx86 with Intel TDX or AMD SEVCode execution where the buyer's guarantee is that the host cannot read the workload. Without the CPU feature there is nothing to attest, and the attestation is the product.
Open-model inferenceNVIDIA GPU, 24 GB VRAM or moreServing open-weights models. Below that the models worth serving don't sit resident in memory, and a model that has to be paged isn't worth scheduling.
Not a fit

Consumer CPU boxes are out.

Spare desktops, laptops, home servers, general-purpose CPU-only machines: not on the list, and we'd rather say so on the landing page than in a rejection email. We ran the arithmetic both ways and it closes on neither side: no price is simultaneously low enough that we'd prefer it to commodity cloud CPU and high enough to be worth your hardware and your electricity.

So: never buy hardware for Sable. Not for this class, not for the other two. Nothing on this page is a reason to spend money.

The contract

Read what a node has to do first.

The wire contract a node speaks is published, the same HTTP execution shape Sable's own sandbox path already uses. It is deliberately out ahead of the programme: before there is anything to join, you can read exactly what a machine would be asked to accept, run, and report, and decide it isn't worth it.

It is also only the execution call. Enrolment, heartbeats, and scheduling now run gateway-side (Sable's own node is enrolled through them), but they aren't part of the published contract yet. Publishing the part a third party would speak first is how the rest stays honest as it opens.

Node agent · wire contract
What this page is not

Three things we're not saying.

01

Not an earnings offer.

No rate, no revenue share, no payback period, no utilization estimate. We haven't set them. Anyone who tells you what a machine will earn on a network that hasn't opened is guessing at your expense.

02

Not a date.

Enrolment opens when there is real demand for the compute and the trust, abuse, and payout machinery to stand behind it. Naming a quarter here would be a guess dressed as a commitment.

03

Not a supply side that exists.

Zero third-party machines serve Sable traffic. Nobody is earning anything from Sable compute, because there is nobody to earn it. If anything on this site reads otherwise, that's a bug; tell us and we'll fix the copy.

Register interest

There's no form. There's an inbox.

We haven't built a signup flow, because a signup flow would imply there's something to sign up for. Email us instead and we'll keep the note.

  • What hardware you run or could get access to: CPU model and whether TDX/SEV is available, or which GPUs and how much VRAM.
  • Where it physically sits, and whether it's a datacenter, an office, or a house.
  • Connectivity: a public endpoint, or behind NAT.
  • How many machines, and whether they're yours or rented.

What you won't get back is a rate card, because there isn't one. What you will get is a straight answer about where this is.