Isn't this just more expensive than a normal VPS?

Per gigabyte of RAM, yes — a mainstream host with a card and ID check will be cheaper. The Pro line isn't competing on that number. It exists for people who need a lot of RAM AND can't (or won't) hand over a card and passport. If KYC isn't a problem for you, buy the cheap one — we mean that.

How much RAM does a serious AI workload actually need?

Depends what you run. A multi-agent crew that only calls an LLM API is light. But hold a few million embeddings in memory for fast RAG, keep long conversation history, or run several services on one box, and you're past 32 GB quickly. Local model inference on CPU eats even more. That's the zone Pro is built for.

Can I run a local LLM on this?

Small quantized models on CPU, yes — slowly. Pro plans are CPU + RAM, no GPU. For real-time local inference you want a GPU host. Most people use the RAM for orchestration, vector search and memory, and call an API for the heavy generation.

What do I actually get that a cheap host doesn't?

High RAM without an identity check, crypto payment, a dedicated IPv4 with all ports open, nightly auto-backups, and the whole thing orderable by an AI agent over MCP or REST. It's the combination that's rare, not any single piece.

Which crypto and networks, and is there really no KYC?

USDC or USDT on Base, Ethereum or Polygon. Email signup only — no card, no identity document. No-KYC means we don't ask who you are; it doesn't mean anonymous (we still see the server's traffic like any host).

Why a no-KYC high-memory VPS costs more than the $/GB math suggests

If you go looking for a 64 or 80 GB VPS and you land on our Pro line, the first thing you'll probably do is divide the price by the RAM. Then you'll open a mainstream host in another tab, do the same math, and think: these guys are charging a premium for the same gigabytes.

You're not wrong about the arithmetic. Let me tell you why the arithmetic is the wrong lens here.

The honest part first

Per gigabyte of RAM, a big commercial host is cheaper than us. Sometimes a lot cheaper. If your only requirement is "I need a box with a lot of memory and I don't care about anything else," you should go buy that box. I'm not going to pretend a Pro plan wins a pure $/GB race, because it doesn't, and you'd catch me lying inside five minutes.

So if that's you — go. Seriously. No hard feelings.

Still here? Then you probably have a second requirement, and it changes everything.

Try to actually buy it

Go and try to rent 80 GB of RAM without a credit card and without uploading an ID. Not "sign up and add a card you promise not to use" — actually pay with crypto, with nothing but an email, and get root.

You'll find that almost nobody sells that. The cheap per-GB hosts are cheap partly because they run you through payment rails and identity checks that filter out exactly the customers who want to pay in USDC and not say who they are. The moment you need real memory and no KYC, the field of options collapses to a handful — and the per-GB price stops being comparable, because you're no longer shopping for the same product.

That's the whole point. Pro isn't "expensive RAM." It's "RAM you can get on terms almost nobody offers."

What the price is actually for

When you pay for a Pro plan you're buying a bundle that normally doesn't come together:

High memory — 32 to 80 GB, enough to hold a real vector index in RAM or run a fleet of agents that share state.
No identity check — email and crypto, that's it.
A dedicated IPv4 with every port open, so you can actually host something that listens (an API, a model server, a VPN endpoint), not just make outbound calls.
Nightly auto-backups on the Pro line, so a 64 GB box full of your work isn't one bad command away from gone.
Agent-orderable — an AI agent can provision it over MCP or REST without a human in the loop.

Any one of those you can find somewhere. The combination, paid in crypto, is the rare thing.

Who this is genuinely for

A few workloads eat memory for real reasons, not because someone over-provisioned:

Self-hosted RAG at scale. Good vector search wants the index in RAM. A few hundred thousand embeddings fit in a couple of gigs; a few million, with the app and OS around them, and you're looking at 32–48 GB before you've done anything clever. Keep it on disk instead and search gets slow — the whole reason you self-hosted evaporates.

Multi-agent systems that hold state. A crew of agents that only fire off API calls is light. But give them shared memory, long histories, a working set they pass around, and the RAM curve bends up fast. This is where a single 64 GB box quietly replaces three smaller ones you'd otherwise be juggling.

Private inference and data work. Running a quantized model on CPU, or crunching a dataset that has to stay in memory, or co-locating several production services — memory-bound, all of it. And if the data is sensitive enough that you're paying in crypto to avoid a paper trail, a server you fully control is the point, not a nice-to-have.

When you should not buy Pro

Let me save you money. If KYC genuinely doesn't bother you and you just want cheap RAM — a mainstream host wins, take it. If your agents only call an LLM API and barely touch local memory, our own NAT and dedicated-IP plans from $3–20 are plenty; you don't need Pro. And if you need a GPU for real-time inference, no CPU-and-RAM box (ours or anyone's) is the answer.

Pro earns its price in exactly one situation: you need a lot of memory and you need to pay privately, with no card and no ID. When both are true, the per-GB comparison isn't just unfavorable to us — it's measuring the wrong thing, because the cheap option was never actually available to you.

That's the plan. If it fits, the Pro line starts at 32 GB; size up from there.