Billing

predictAI bills in one currency: credits. Compute-heavy actions cost credits, charged when the work runs. Your plan grants a monthly credit allowance, caps how many objects you can hold, and unlocks capabilities. Your balance, rates, and quota usage are all readable by API:

Plan    = monthly fee + included credits + quotas + capabilities
Balance = one credit wallet per account (user or organization)
Actions → debit the wallet;  grants and refunds → credit it back
Quotas  → cap object counts PER WORKSPACE (pipelines, models,
          deployments, goals, …); only seats & workspaces are account-level

If an action would cost more than you have, it returns 402 and simply doesn't run — nothing is half-charged. If creating an object would exceed a quota, it returns 403 with a message naming the limit.

Thirty seconds of API

# Your credit balance
curl "$API_BASE/v1/billing/balance" \
  -H "Authorization: Bearer $TOKEN" \
  -H "X-Workspace-Id: $WORKSPACE_ID"

# Price an operation before you run it
curl -X POST "$API_BASE/v1/billing/estimate" \
  -H "Authorization: Bearer $TOKEN" \
  -H "X-Workspace-Id: $WORKSPACE_ID" \
  -H "Content-Type: application/json" \
  -d '{ "operation": "adapter_training.specialize", "units": 1 }'

In this section

Plan upgrades, credit packs, checkout, and invoices are managed in the app, not through the API.

The wallet is account-level — charges land on your personal wallet, or on your organization's wallet when you operate in an organization context — but quotas and included monthly allowances are per workspace: every workspace on your plan gets its own full allowance. What each action costs at the moment it runs is documented where the action lives — Trainings, Inference, Goals, and BYOM uploads all link back here for rates.