Prerequisites
Kumo containers will be publicly available in the NVIDIA NGC Registry soon. Stay tuned.
These requirements apply to both the Helm and Docker deployment paths. Satisfy them before following either path’s setup guide.
NGC access
Kumo container images are distributed through the NVIDIA GPU Cloud (NGC) private registry.
You need an NGC API key with access to the 1000161941370113/kumo-3 organization, listed in NGC as Structured Data & Graph Models / kumo-3.
Credentials are provided by your NVIDIA or Kumo contact.
There is no self-service sign-up path.
The NGC catalog is available at catalog.ngc.nvidia.com. A public registry is coming soon and will cover both deployment paths.
Once you have your key, authenticate Docker against the registry:
Single-quote '$oauthtoken' so your shell does not expand it.
This authentication is required for Docker Compose image pulls.
For Helm, you also use the same key to fetch the chart from NGC (refer to Registry & Image Access).
Tenant identifier
For Docker Compose, set CUSTOMER_ID to a short, lowercase name that identifies your Kumo deployment.
For Helm, set global.tenant to the tenant name.
Helm uses this value in storage paths, pod labels, and the default Temporal namespace.
Example: acme.
Image tag
All three Kumo images (kumo-backend-cpu, kumo-backend-gpu, kumo-ui) share a single image tag.
The tag is a 40-character git commit SHA - not latest or a semver alias.
Confirm the exact tag to use with your NVIDIA contact before starting either deployment path.