Core concepts
The forecasting loop
How the pieces fit together, end to end — with the API call for every step.
Every production forecasting system on predictAI is the same loop. Here it is with the actual endpoints:
┌────────────┐ ┌─────────┐ ┌────────┐ ┌──────────┐ ┌────────────┐ ┌───────────┐
│ Ingest │ → │ Signals │ → │ Model │ → │ Pipeline │ → │ Deployment │ → │ Inference │
└────────────┘ └─────────┘ └────────┘ └──────────┘ └────────────┘ └───────────┘
sources, jobs push, list catalog, segment + promote the one URL,
files, webhooks analyze foundation schedule best training predictions| Step | What you do | Endpoint |
|---|---|---|
| 1. Ingest | Connect a source, upload a file, or push values | POST /v1/signal/push, Ingestion |
| 2. Curate | Build a segment over the signals a model should see | POST /v1/segment |
| 3. Model | Pick a foundation model or define your own | /v1/models |
| 4. Pipeline | Bind segment + models + schedule | POST /v1/pipelines |
| 5. Train | Wait for the schedule, or trigger now | POST /v1/pipelines/{id}/train |
| 6. Deploy | Put the best training live | POST /v1/deployments |
| 7. Forecast | Call the pipeline's inference alias | POST /v1/inference/pipelines/{id}/infer-segment |
The shortcut
Goals collapse steps 2–6 into one call:
| Step | What you do | Endpoint |
|---|---|---|
| 1. Ingest | Same as above | POST /v1/signal/push |
| 2. Goal | Declare the outcome | POST /v1/goals |
| 3. Forecast | Call the goal's inference alias | POST /v1/inference/goals/{id}/infer-segment |
Staying in the loop
Two supporting systems watch the loop for you:
- Notifications — the feed, delivery channels, and preferences for platform events.
- Realtime — WebSocket streams of training progress, deployment changes, and goal alerts, so you never poll.
And one system meters it: Billing — trainings, inference, and discovery runs consume credits; quotas cap how many of each object your plan allows.