Preview and availability
A precise description of what this developer preview does and does not promise.
Current product surface
Oiy AI is a container-level elastic GPU service with a Docker-like workflow, resource resizing, scale-to-zero through idle sleep, wake on demand, and usage-based billing. Supporting capabilities include persistent storage, templates, authenticated endpoints, task protection, API keys, account-scoped private-image access, and developer interfaces.
Implementation is not a guarantee that a particular placement is enabled. Live deployment depends on regional configuration, independently configured prices, available hardware, a connected runtime, and account eligibility.
Supported profiles versus live inventory
The GPU catalog defines PRO 6000 Blackwell, H200, B200, and B300 profiles. A profile must have matching inventory and pricing before it can be selected for a real allocation. A blank price is unavailable, not free. Model names do not imply equivalent hardware is silently substituted.
Use the console or GET /api/catalog for current catalog information. Catalog configuration is not a reserved allocation; available capacity can change before a create request is accepted.
Developer distribution
The HTTP API is the common interface. Python SDK/CLI and MCP implementations are provided to preview users through a source checkout. Do not assume pip install oiy-ai or npx @oiy-ai/mcp is a published installation path. Follow the specific developer setup guides.
Outside this preview
- Function/decorator deployment APIs and automatic horizontal replica autoscaling. Container-level scale-to-zero and explicit resource resizing are implemented.
- Managed multi-node training clusters.
- Public anonymous inference endpoints or a general TCP/UDP port-management interface.
- Automatic migration of an existing volume between placements.
- Guaranteed zero cold start, benchmark throughput, or a stated uptime SLA.
These are product boundaries, not release promises. Do not design against an unannounced roadmap.
Protect your work
Keep your own backups of valuable data. Read deletion and billing behavior before removing resources. When a lifecycle request returns an uncertain result, inspect current state before sending another mutation.