Cradle as a datacenter service
Don't rent GPUs — sell verified results. Customers hand over a task and data; Cradle's agents work under a two-tier risk system, an operator approves, and the customer gets a verified answer with citations and a full audit trail.
Three delivery models
Pick the one that fits the customer's perimeter — the same bootstrap flow serves all three.
Curl installer
For open and self-hosted cases with internet access. A one-line install.sh pulls the bundle from S3 and rebuilds native modules on the host.
Self-contained tarball
For closed networks with no registry. A prebuilt archive ships the bundled server plus native modules compiled for the target OS and GPU — fully offline after delivery.
VM image
The datacenter keeps a golden image with Cradle Server preinstalled. The customer downloads only the client and pastes in credentials — no server-side install step.
Topology progression
Start simple, grow only when you need to.
Everything on one server
cradle-server plus a local CUDA runner on a GPU node. The client is a thin console over the API. Brain and inference live together — the least new code. Recommended starting point.
Control plane + data plane split
Cradle manages an inference daemon on a separate GPU box: lifecycle and model selection over a control channel, inference traffic over HTTP via a remote-gpu provider.
Shared GPU pool + queue
Scaling, spot instances, and automatic task distribution across a pool. The advanced end of the spectrum — not something to build until T0 works.
Hard tenant isolation
Every tenant gets its own infrastructure and its own database — tenants never share a perimeter. Isolation is instance-per-tenant: a separate cradle-server with its own data directory, database, and perimeter. A shared instance with "project = tenant" is explicitly ruled out.
- projectId partitions agents, knowledge bases, tickets, and budgets inside a tenant — never between tenants.
- Embedded SQLite means data is literally one file in the tenant's data directory; nothing leaves the perimeter.
- Choose your isolation level: process per tenant, container per tenant, or a full VM per tenant.
- A datacenter-side orchestrator provisions, starts, stops, and tears down tenant instances and hands credentials to the customer portal.
Offer Cradle to your customers
Talk to us about a managed deployment, or read the deployment docs to see the mechanics.