Credits and limits

Two things are metered on the platform: AI credits and storage bytes. They are capped separately, refused separately, and — importantly — scoped differently. Everything below applies identically whether you call the API with a dashboard session or with a VIITE_API_KEY.

The tiers

	Free	Pro (€25/mo)
Seats	1	5 (more if you buy more)
AI credits	5,000 one-off	10,000 one-off
Storage	100 MB	1 GB

The AI allowance is one-off, not monthly — it is a grant, not a refill. Every user starts with 5,000 credits; activating Pro tops the subscribing owner up by another 5,000. When it runs out you buy top-ups (10,000 / 25,000 / 100,000-credit packs at €10 / €25 / €100 — €1 buys 1,000 credits) rather than waiting for a reset. Storage is a standing ceiling: delete assets and the space comes back.

What spends a credit

Anything that calls a language or speech model on the platform’s account:

• the AI job steps (agent, enrich, analyze, triage, implement_task, tts);
• the social and blog composers (createSocialFromLink, createBlogFromLink, createBlogFromPostDraft);
• text-to-speech synthesis.

Deterministic steps never spend anything, and hybrid steps (watch, collect_invoices) degrade to their deterministic behaviour rather than refusing when there is no allowance left.

Charging is metered on actual token usage with a markup, so a single call typically costs a few to a few hundred credits. Charges are always whole credits, rounded up, with a minimum of 1 — your balance and every amount you read back is a whole number. Read your balance any time:

const credits = await client.credits.getMyBalance();

Scoping: credits are per-user, storage is per-org

This trips people up, so it is worth stating plainly:

• Storage is scoped to the organization. Every member’s uploads count against the same ceiling.
• Credits are scoped to the user. An API key spends the balance of the admin who minted it — not a shared org pool.

So two keys minted by the same admin draw on one balance, while keys minted by two different admins draw on two. If you want a predictable meter for a server-side integration, mint its key from a single dedicated admin account — and note that a Pro org’s extra grant lands on the owner’s balance, so a key minted by a different admin sees only that admin’s own starting 5,000.

Being refused: HTTP 402

Both limits refuse with 402 Payment Required and a message naming the limit you hit. Out of credits:

Out of credits — your plan’s one-time AI allowance is used up. Buy more credits or upgrade at /upgrade.

Over the storage ceiling:

Storage limit reached: this plan allows 100.0 MB, and this organization is already using 101.0 MB. Upgrade for more storage at /upgrade.

The refusal happens before the expensive work: no model is called and no bytes are written, so a 402 costs you nothing and is safe to retry after topping up. Handle it by status, not by string:

import { ApiRequestError } from '@viite-ai/platform-sdk';

try {
  await client.tasks.process(linkId);
} catch (error) {
  if (error instanceof ApiRequestError && error.status === 402) {
    // Top up, then retry — nothing was consumed.
  }
}

Note

One asymmetry to know about: a job that runs out of credits mid-run cannot answer with a 402, because it is already running in the background. It aborts and records the refusal on the job instead, naming the step that was blocked. Synchronous calls get the 402; background work gets a failed job.

Per-type upload caps

Independently of the quota, uploads are capped per media type (5 MB images, 50 MB audio). A file over its type cap is refused whatever your plan.