Skip to main content
After a deployment is live, the operational work shifts from creation to inspection, validation, and controlled change management.

​
The deployment detail surfaces

The product organizes deployment operations into four tabs:
TabUse it for
OverviewDeployment metadata, public model identifier, and API usage
InstancesRuntime capacity and instance-level status
InferencesRecent requests served by the deployment
SettingsAdministrative actions, including deletion

​
What to monitor

​
Instances

Watch for instance status changes such as:
  • Running
  • Initializing
  • ShuttingDown
  • Terminated
  • failure states
This view is where you confirm that the deployment actually has healthy serving capacity.

​
Inferences

The deployment inferences view is the fastest way to validate:
  • which model identifier is serving traffic
  • whether requests are succeeding
  • token counts
  • request duration
  • total cost
Use it for spot checks after rollout or after any meaningful deployment change.

​
Settings

Settings is where destructive lifecycle actions live. Treat deletion as permanent and use it only when you are sure you no longer need the deployment.

​
Best practices

  • keep a baseline smoke test for every deployment
  • inspect recent inferences immediately after rollout
  • continue routing production traffic through Observe so deployment behavior remains visible in the larger platform analytics
  • avoid making deployment decisions without a current eval baseline

​
Next steps

Observe Overview

Keep production traffic visible after rollout.

Evaluate Overview

Use evals to decide whether a deployment change actually improved the product.