Deployments
Put a trained model into live serving — deploy, monitor, scale, retry, and retire.
A deployment is a training promoted to live serving. Training produces a scored artifact; promoting it creates a deployment record that tracks which pipeline, which training, what serving mode, and whether it's actually ready to answer requests.
Everything lives under one prefix — /v1/deployments — covering both the
resources (list, inspect, delete) and the lifecycle actions (deploy,
retry, unpromote, scale).
Thirty seconds of API
# Deploy: put a completed training into live serving
curl -X POST "$API_BASE/v1/deployments" \
-H "Authorization: Bearer $TOKEN" \
-H "X-Workspace-Id: $WORKSPACE_ID" \
-H "Content-Type: application/json" \
-d '{ "pipeline_id": "'$PIPELINE_ID'", "training_id": "'$TRAINING_ID'" }'
# Watch it come up
curl "$API_BASE/v1/deployments/pipelines/$PIPELINE_ID/status" \
-H "Authorization: Bearer $TOKEN" \
-H "X-Workspace-Id: $WORKSPACE_ID"Once the deployment is ready, run forecasts through
Inference — or subscribe to the workspace's
realtime channel for deployment.updated events
instead of polling.
In this section
Concepts
The promotion flow, Shared vs Dedicated serving, and readiness.
Deploy a trained model
POST /v1/deployments — the request, the response, and every error.
Manage deployments
List, inspect, retry, unpromote, delete, and clean up.
Serving status
Check what's live — per pipeline, per mode, or fleet-wide — and test it.
Scale Dedicated deployments
Set replica counts on dedicated serving capacity.
Trainings themselves live in Trainings; the pipelines that produce them are in Pipelines.