quark · why
We picked different compromises.
Every AI platform asks you to trade something: power for simplicity, control for speed, governance for momentum. Quark's design picks a different set of trade-offs — here's what makes it worth a look.
the base layer
Products on a foundation, not products on sand.
Most “AI platforms” are a thin product floating on somebody else's infrastructure, with a wall between what you can see and what actually runs. Quark is the opposite: the AI products — AI Gateway, Sandboxes, Workflows, Knowledge Graphs, Vector Indexes — sit on a foundation that is itself first-class and inspectable.
The gateway rides on real identity and real secrets. A sandbox is a real, observable workload. Retrieval pipelines read data that stays in your custody. Nothing is a simulation, and nothing is hidden — when the AI layer does something, the foundation underneath is the same one you can inspect, govern, and reuse.
the deployment model
Foundation always on. AI by selection.
A Quark deployment ships complete: the Foundation — identity, access, secrets, metering, quotas — always on, and only the AI products you selected. Didn't select Vector Indexes? The console doesn't offer it. Fewer products never reads as a broken platform; it reads as a platform that knows itself.
Whatever a selected product needs comes with it — activated, visible, shared if it's already on. Nav, dashboard tiles, and pages all follow the selection. What you see is what your deployment does.
your terms
Your deployment, your terms.
The same platform runs as a hosted service or on hardware you own, and the same console runs in the browser or on the desktop.
Point the console at a hosted deployment and work; point it at a platform running on your own metal — a lab rack, an on-prem cluster — and the experience doesn't change. Build against your own deployment, then promote the identical workloads to a shared one, with nothing to move.
on purpose
A modern base, on purpose.
Under the product surface, Quark is built the way infrastructure is built when it's meant to be trusted.
From the platform's own services down to the shared primitives — a single, memory-safe language across the stack, with supervisor-style restarts and standard health probes in the base.
The platform speaks a versioned, strongly-typed API with machine-readable schemas — not a wall of bespoke REST. SDKs and a CLI come from the same contracts the services use.
Resources are named for what they are — sandboxes, indexes, routes, runs — not for a sales category. Your workloads and your knowledge of the platform aren't chained to a catalog.
Multi-step changes are recorded, resumable operations; health, readiness, and metrics endpoints are standard on every service.
safety
Safety conventions, everywhere, by default.
The details that separate a platform you tolerate from one you trust.
API keys and secret values display exactly once, with a deliberate confirmation — never again.
Workflow versions, index versions, published endpoints — history doesn't mutate.
Destroying a sandbox takes a typed confirmation; lifecycle history is append-only.
Explicit deny rules veto role grants, everywhere.
Secret operations and platform activity leave readable trails.
one vocabulary
One console, one vocabulary.
From the front page to the deepest detail pane, the platform speaks one language: Routes, Sandboxes, Workflows, Indexes, Runs, Spend. The words you learn reading this site are the words you'll see in the console, in the API, and in the audit log.
Onboarding is shorter because there's nothing to untranslate.
contact
Talk to us about running Quark.
Tell us about your workloads — we will show you what your own cloud can do.