Free to light. Cheap to scale.
You already bought the GPUs. Clystria just makes them one computer. Start free on the hardware under your desk; pay only when your cluster turns into a business.
Spark
Open-coreFor tinkerers turning the machines under their desk into one.
Install the agent- ✦Up to 4 nodes
- ✦Unlimited models & deployments
- ✦Unified VRAM pool
- ✦Automatic layer placement
- ✦Self-healing failover
- ✦OpenAI-compatible endpoint
- ✦Community support
Foundry
For teamsFor teams running a private AI cloud they actually depend on.
Start free trial- ✦Everything in Spark
- ✦Unlimited nodes
- ✦Autoscaling & rolling updates
- ✦Multi-cluster management
- ✦Live observability & tracing
- ✦Role-based access control
- ✦Priority support · 8×5
Smelter
EnterpriseFor orgs forging clusters across racks, buildings, and regions.
Contact sales- ✦Everything in Foundry
- ✦Air-gapped & on-prem deploys
- ✦SSO / SAML / SCIM
- ✦Guaranteed uptime SLA
- ✦Dedicated solutions engineer
- ✦Custom hardware profiles
- ✦24×7 support
Questions, answered
Is it really free?
The Spark tier is open-core and free forever for up to four nodes. No credit card, no time bomb. You only pay when you scale past a homelab.
What does 'per node' mean?
A node is any machine running the Clystria agent, regardless of how many GPUs it has. A box with eight H100s counts as one node.
Does my data leave my network?
Never. Clystria runs entirely on your hardware. Model weights and every request token stay inside your building. Billing telemetry is metadata only, and air-gapped licensing is available on Smelter.
Can I mix consumer and datacenter GPUs?
That's the whole point. A 4090, an A6000 and an H100 can share one pool — Clystria weighs each by its measured throughput.