Pradra StudioDocs
Sign in

Deployments

A model only creates value once your applications can call it. A deployment is a running instance of a capability's production model, reachable as a stable, contract-checked endpoint. You create it once; the platform keeps it running and lets you update or stop it.

The lifecycle

A deployment moves through: Created → Starting → Running → Updating → Stopped → Deleted.

Two properties make deployments boring in the best way:

  • The endpoint never changes. Promoting a better model, or rolling back to the previous one, is a pointer switch behind the same URL. Your applications never redeploy because the model improved.
  • A deployment refuses to start broken. Before starting, the platform cross-checks the capability's input contract against its feature set: if a derived feature reads a column the contract doesn't declare, the deployment fails fast with a clear message — instead of serving a model whose inputs are quietly wrong while the endpoint looks healthy.

Working with deployments in the Studio

  1. Go to Deployments in the sidebar and create one: pick the solution, the capability, and the production model version to serve.
  2. Watch it move from Starting to Running — the list polls live.
  3. Open the deployment's detail page. It is the operational home for the capability:
    • the endpoint your application calls, ready to copy;
    • a try-it playground — send a test prediction from the browser and see the shaped output plus round-trip latency;
    • monitoring charts (requests, errors, latency percentiles);
    • drift status for the incoming data;
    • batch jobs submitted against this capability;
    • prediction history — the audited record of every call.
  4. Stop a deployment to take it offline without deleting its history; delete only when the capability is truly retired (confirmed, like every destructive action).

Prediction history

Every prediction — real-time or batch — is recorded in prediction history: what was asked, what was answered, when, and how fast. For generative capabilities, the record also carries token usage and cost. History gives you an audit trail and feeds monitoring; it is kept for a configurable retention window.

Next