# Limits and routing

> Monthly limits in USD for each key, member and customer; the three routing modes; and what happens when a limit is reached.

Source: https://kongapiai.com/en/docs/limits

## How limits work

Limits are monthly, in USD, and reset on the 1st of each month, UTC.

You can set a limit at several levels:

| Level | What it caps |
| --- | --- |
| Key | The usage of one key |
| Member | The total usage of all keys that belong to one person |
| Customer | The total usage attributed to one of your customers |
| Credits | The credits left in the whole workspace |

A request stops at the first limit it reaches. The API returns `402` with a code that names the limit that ran out, such as `key_limit` or `budget_exceeded`.

> Before it runs, each request holds an amount equal to the most it could cost. If the available credits cannot cover that hold, the request is refused with `insufficient_budget`. When the request ends, the unused part is released.

## What a limit counts

A limit counts only what is paid with KONG API AI credits. Requests served by your own AI sources are not capped by these limits; their limit is the quota your provider sets on your account.

## Three routing modes

Each key runs in one of three modes:

| Mode | Where requests run | KONG API AI token charge |
| --- | --- | --- |
| Auto | Your own sources first. Credits only if you turned on fallback | Only when credits are used |
| Credits | Always paid from credits, at the KONG API AI price | Yes |
| Own sources | Only your sources | No |

In Auto mode, if none of your own sources can serve and fallback is off, the API returns `429` with `own_quota_unavailable`.

## When a limit runs out

- Raise the limit of the key, member or customer in the console.
- Or add credits.
- The next request goes through. You do not need a new key.

## Pause all keys

An admin can pause every key of the workspace on the Billing page. The API then returns `503` with `workspace_paused`.
