---
title: 'Credits and model access'
description: 'See how requests spend credits, read your balance, buy credits, and set department spending limits.'
url: 'https://docs.streamtoken.dev/credits'
last_updated: '2026-11-11'
---

> Docs index: https://docs.streamtoken.dev/llms.txt · MCP: https://docs.streamtoken.dev/mcp · Skill: https://docs.streamtoken.dev/skill.md

# See where your credits go

Credits pay for model requests that Token Streamer runs for you. Open **Credits** in your organization workspace to see your balance, model usage and reference prices. Owners and administrators can see this page.

Requests made with your own provider key ([BYOK](https://docs.streamtoken.dev/providers.md)) do not spend credits. Your provider bills you for them.

## Model usage and spending

The top of Credits shows model usage and estimated provider spending in the currently selected inference region. Select a model, include or exclude model tests, and choose **Day** or **Week** with the reporting-date picker. Day shows 24 UTC hourly intervals. Week shows seven UTC dates ending on the selected date. Future dates are capped at today; invalid dates use today.

**Model usage** switches between requests and recorded tokens. **Provider spend estimate** sums only requests with recorded usage and reference rates. The coverage count shows how many matching requests have estimates. A dash or chart gap means no estimate, while an explicitly recorded zero remains zero. Each chart includes a **View interval data** table for exact values and keyboard/screen-reader access. The model table breaks the selected period down by model.

Your provider bills requests made with your own key. These estimates are not invoices or actual billed costs: they exclude cache adjustments, account-specific discounts and requests without usable usage/rate metadata. Requests made with your own provider key have no Token Streamer usage charge. These estimates do not spend your credits.

Charts apply the same organization, inference-region and metadata-retention scope as Usage. The credit balance below is a separate organization-wide total; it is not split by inference region.

Estimates are recorded in picos (`provider_estimate_picos`, one trillionth of a USD). To convert to USD, divide by 1,000,000,000,000.

## Buy credits

Choose **Billing** from the user menu, select a credit pack, and select the checkout button. Owners and administrators can buy credits. A secure hosted payment form opens. The purchase stays with the organization that started it, even if you switch workspaces during checkout.

Checkout adds VAT where it applies. VAT does not add credits: you receive the credits of the pack, not of the VAT-inclusive total.

Returning from checkout does not add credits by itself. **Awaiting payment confirmation** means the payment is not confirmed yet. Credits are added once, after the payment is confirmed. **Paid · awaiting gateway** means the credits are recorded and are about to become available for requests. Refresh the Credits page to see the latest state.

If checkout was interrupted, choose **Resume checkout** to continue the same purchase instead of starting a second one. A failed or expired payment adds no credits. After a refund or dispute, new spending from the credit pauses while we review it. Requests already in progress can still finish. Contact support if your balance is paused.

Pro is a separate monthly subscription and does not add credits. See [Pro subscriptions](https://docs.streamtoken.dev/subscriptions.md).

## Department spending limits

A department limit caps how much credit its keys can reserve and spend. It covers every credit in the organization wallet. It is checked together with the organization wallet, so neither can be exceeded, even with parallel requests. Organization keys do not count against a department limit, and a department key is never rerouted through an organization key.

An owner or admin chooses when the limit resets on the **Departments** page:

| Setting | Resets at |
| --- | --- |
| Never resets | Never. The limit is a total cap. |
| Daily | 00:00 UTC |
| Weekly | Monday 00:00 UTC |
| Monthly | The 1st of the month, 00:00 UTC |

At a reset, settled spend starts again from zero. Requests still in progress keep their reservation and settle into the new period. If you change the period, the spend counted so far stays until the next reset, so the change never frees budget early. Each change of limit or period gets a new revision and an audit record. The **Departments** page lists the last 20 changes for each department: the old and new limit and period, who made the change, and when. The highest limit is 1,000 credits.

Department spend counts settled usage only; a reversal after a refund or dispute does not change department spend. The Credits page shows each department's spend in its current period, and its pending amount: requests from the department's keys that are waiting for reconciliation. A pending request still holds its reservation against the department limit until it settles or is released. Requests still in progress for a few seconds are not listed per department; they show only in the organization's **Reserved for requests** total.

| Status | Code | Meaning |
| --- | --- | --- |
| 402 | `department_spending_limit` | The department does not have enough limit left for this request's maximum reservation. The message states the remaining credits, the limit and the reservation. Lower `max_completion_tokens`, or ask an owner or admin to raise the limit. |
| 503 | `department_policy_stale` | The department limit changed while the request was being admitted. Retry the request. |

## Read your balance

- **Posted balance** is the balance after settled requests.
- **Available after reported holds** subtracts the reservations the gateway has reported and credits it has not confirmed yet.
- **Reserved for requests** is the maximum amount held for requests that have not finished.
- **Awaiting an outcome** is the part of the reservation whose provider outcome is uncertain.
- **Measured usage** is the total charge settled from observed token counts.

Balances show the latest report and its time, so they can be slightly behind during an outage. The gateway enforces your limits itself, so the page is not permission to spend.

## How a request spends credits

Before a request runs, the gateway holds the maximum amount the request could cost. The hold can be much larger than the final charge. When a valid response arrives, the measured usage is charged once and the unused hold is released. Credits are shared by all of the organization's keys. The price shown in the workspace is a reference price for the model; cached input is priced separately from uncached input.

If a request is rejected for lack of credit, lower `max_completion_tokens`, add credits, or ask an owner to raise the limit.

### Retry without repeating a model call

Use a distinct `Idempotency-Key` for each intended request. It must contain 16-128 letters, digits, underscores, or hyphens. Reusing a key returns `409 request_exists`. It does not replay the original answer or send another provider request. Prompt and completion text are not kept in the accounting records.

If a response says `sync_pending` in `X-StreamToken-Accounting`, your usage is safely recorded and will appear in your balance shortly. The model request is not repeated.

### Uncertain outcomes

A network failure, provider server error, or missing usage after a request is sent leaves the hold awaiting review. You are not charged an estimate and the hold is not refunded automatically. If the provider clearly rejects the request, the hold is released.

Error responses use `reconciliation_pending` for uncertain outcomes. Keep the operation ID from the response (`X-StreamToken-Operation-Id`). Pending holds stay reserved until usage is measured or the outcome is confirmed.

**Development credit operations** in [Diagnostics](https://docs.streamtoken.dev/diagnostics.md) lists recent outcomes, including completed usage, released holds, and requests needing review. Use **Pending only** to find holds awaiting review, or enter an **Operation ID** to look up one request. If you need a pending hold resolved, contact support with the operation ID.

## Limits

Credits are used only on supported models, within your key permissions and any department limit. The gateway refuses new work when an internal limit is reached rather than risk overspending.