Configure and Validate#

Validate the deployed service before onboarding users. A healthy /health response is the first check; a packaged, downloadable POC and a restart check establish the user workflow and durable state.

Prerequisites#

Gather the deployed URL, credentials, and runtime access needed for the checks.

  1. Complete the Docker Compose or Kubernetes deployment.

  2. Obtain the browser URL and access to Settings, service health, and deployment logs.

  3. Have an inference profile and, for the default auto generation path, a valid Cursor credential.

Validate Inference and Identity#

Check the browser and provider configuration before submitting a generation job.

  1. Open the browser-facing URL over HTTPS for a shared deployment. Confirm sign-in completes and returns to the application without a redirect loop.

  2. Confirm the exact origin, host, and OAuth callback values match the external URL. Shared deployments must use secure cookies and managed signing, OAuth, database, inference, and encryption secrets.

  3. In Settings, check the effective inference profile. An administrator can provide the base URL, key, and five MODEL_* roles, or users can save an OpenAI-compatible profile when overrides are enabled.

  4. Validate chat roles through /chat/completions and the embedding role through /embeddings. A /models response alone does not prove the profile can run a workflow.

  5. Validate the Cursor key if CODE_GENERATOR_TOOL=auto or cursor is selected. Choose llm only for the explicit inference-backed full-project mode.

Run the Deployment Acceptance Test#

Record the observed result of each check for the deployment handoff.

  1. Confirm the three POC Factory images resolve to 1.1.0, or approved digests from the same release, for the host CPU architecture.

  2. Confirm required pods or containers are Ready, then query /health and check database and state-manager results.

  3. Confirm the frontend loads and /api requests reach the backend. For Kubernetes, inspect ingress behavior for long-running Server-Sent Events.

  4. Generate one small POC using Generate your first POC. Confirm it reaches packaging, remains on Dashboard, and downloads.

  5. Record each validation gate as passed, failed, or skipped. A skipped Docker build or runtime gate is not a container validation pass.

  6. Restart the backend or Compose stack and confirm the POC record, archive, job state, and saved settings remain usable.

  7. Check that logs identify the job; check Phoenix or platform telemetry when enabled.

Deployment

v1.1.0 default candidate validation

Meaning of a successful run

NGC Docker Compose

Static and test checks plus Docker build, container start, and runtime checks, subject to a working host Docker daemon

The configured candidate gates ran and passed

Kubernetes production values

Static and test checks; Docker-dependent deployment and runtime checks disabled

The configured in-cluster gates ran and passed; generated-container build and runtime remain unproven

The Kubernetes chart rejects runtime validation without deployment validation and deployment validation without Docker socket access. Use a separate validation environment if production cluster policy excludes socket access.

Record the Handoff#

Record the external URL, release image digests, Helm values or Compose revision, Secret and certificate owners, database and PVC backup policy, retention period, enabled optional services, validation flags, skipped gates, support contact, and acceptance date. Do not record secret values.

Next Steps#

Use the operating guide after the deployment passes acceptance.