quark · infrastructure platform for ai workloads

Run AI workloads on infrastructure you own.

Quark gives your team its own platform for AI work — model routing, sandboxes, workflows, and retrieval, on one foundation of identity, secrets, and metering. Hosted, or on hardware you own. One console, one API.

Contact sales See what's inside

Hosted, or on hardware you own

Same setup, either way

Open tools compatible

No lock-in, leave anytime

how quark deploys

Foundation always on. AI by selection.

Foundation

Identity & keys, Access, Secrets, Metering, Quotas — in every deployment, never an upsell. Every product runs on them, and nothing on the platform goes unwatched.

AI Gateway, Sandboxes, Workflows, Knowledge Graphs, Vector Indexes — your selection, from a single product to all of them.

Whatever a product needs comes with it — activated and visible in your console. Growing is a toggle, not a migration.

capabilities

One platform, everything you run.

The AI layer and the foundation under it, as one system — enable one product or all of them. The console only shows what's on.

Models behind one door

Every model call routed through the gateway — fallback chains, live serving checks, and per-key limits, with usage attributed to the key that made it.

Untrusted work, contained

Sandboxes give generated code, evaluations, and batch work real compute with a TTL — destroyed on schedule or on demand, never holding credentials they weren't granted.

Your data, queryable

Knowledge Graphs and Vector Indexes turn your documents and records into retrieval you can inspect, rebuild, and wire straight into workflows.

Everything accounted for

Identity, secrets, access, metering, and quotas are always on — every operation watched, bounded, and answered for.

always on

AI Gateway

One endpoint, every model, your rules.

Sandboxes

Ephemeral runs, destroyed on schedule.

Workflows

Step graphs that resume, not restart.

Knowledge Graphs

Your data as queryable relations.

Vector Indexes

Similarity search, declared like code.

Identity & keys

Orgs, projects, keys — shown once.

Secrets

Sealed values with named grants.

Access

Roles and policies; a deny always wins.

Metering

Usage attributed per key, per product.

Quotas

Caps enforced where usage lands.

All of it runs inside your cloud — nothing leaves your network.

Explore all products See what teams run on it

how it works

From host to first run in four steps.

One installer, one sign-up, one first call. This is the whole path.

Point the installer at hardware

Yours or ours. One command against your first host — or nothing to install if you start hosted. The platform registers itself, and the Foundation is on from the first second.

Make your key

Create the org, the project, and your first API key. It's shown exactly once — then it just works, scoped to the project, from the very first call.

Run something real

Call a model through the gateway, drop an untrusted job into a sandbox, or check a workflow into versioned history. All three sit on the same identity and the same metering.

Watch usage answer for itself

Every run lands on a timeline; every unit is attributed to a key and a product. Quotas enforce caps where usage lands — no mystery at the end of the month.

Follow the full path

safety by default

Trust is the default.

None of this is a paid tier, a flag, or a discipline your team has to remember. It's how the platform behaves when nobody's watching — which is the only time that matters.

Secrets exist exactly onceAPI keys and secret values display once, at creation, behind a deliberate confirmation — then they're sealed. The console itself can't read them back.

History never edits itselfWorkflow versions, index versions, published endpoints — what's published stays. When you roll back, you roll back to bytes, not to someone's recollection.

An explicit deny winsAccess starts from least privilege, and a single matching deny vetoes every grant a role can give. “Never” is enforced, not implied.

Every unit has a name on itUsage is attributed to the key and the product that caused it, and quotas refuse what passes the cap — enforcement happens at the point of use.

faq

Asked by every team.

Do we need to replace our tools?

No. Quark works with the tools you already use — same APIs, same setup, no rewrites on day one.

Where does it run?

As a hosted service, or on hardware you own — your servers, your racks, your network. The same console and the same workloads either way.

What if something breaks?

Quark is built to keep running through failures — and our engineers are one email away when you need a human.

Does our data ever leave our cloud?

Never. Everything runs inside your network, and everything stays encrypted under your control.

Why not just use the big clouds?

You can — until the bill, the lock-in, or the compliance audit lands. Quark gives you cloud-grade infrastructure you own outright. See the comparison

What does it cost to start?

Your bill follows your selection: you pay for the AI products you enable, and the Foundation ships with every deployment. Most teams start with a single product on hardware they already have.

Do we have to deploy everything?

No. Deploy a single product if that's what you need — one gateway, or one sandbox setup — and switch more on when you're ready. Every deployment is the same platform; activation is configuration.

What's always included?

The Foundation: Identity & keys, Access, Secrets, Metering, and Quotas. Every product runs on them, so they ship with every deployment — never as extras.

contact

Talk to us about running Quark.

Tell us about your workloads — we will show you what your own cloud can do.

hello@quarkloop.com