> ## Documentation Index
> Fetch the complete documentation index at: https://docs.nemu.cc/llms.txt
> Use this file to discover all available pages before exploring further.

# Quotas

> Per model spending and usage limits

Your plan already caps requests per minute, per day and per month across the
whole account. Quotas are the layer under that: limits you set yourself, on one
model at a time, so a single expensive model cannot consume everything.

## Rules

A rule is one model, one metric, one number, one reset behaviour.

| Field   | Notes                                   |
| ------- | --------------------------------------- |
| Name    | What the rule is for                    |
| Model   | The model it applies to                 |
| Type    | The metric being counted                |
| Limit   | The number it may not pass              |
| Period  | Rolling or Calendar                     |
| Enabled | Off keeps the rule without enforcing it |

### Metrics

Grouped by window. Daily and monthly reset, total never does.

**Tokens.** Input, output, cached input, cached output, total, cached total.
Available in all three windows.

**Requests, images, audio seconds.** All three windows.

**Spending.** Daily, monthly and total, in euros.

**Searches.** Daily searches and daily deep searches.

### Rolling and Calendar

**Calendar** resets at a fixed boundary. A daily rule goes back to zero at
00:00, a monthly rule on the 1st.

**Rolling** counts a moving window ending right now, 24 hours for a daily rule
and 30 days for a monthly rule, so old usage falls off gradually instead of all
at once.

A `TOTAL_` rule is cumulative and never resets, so the period does not apply.

### Limits

| Free | Starter | Pro | Max | Enterprise |
| ---: | ------: | --: | --: | ---------: |
|    2 |      15 |  50 | 150 |  unlimited |

## Workspace budget

Separate from the rules above, each workspace carries a monthly spend cap. It
defaults to **EUR 10.00**, and setting it to zero means no cap.

The cap is enforced before the request runs, not after. The gateway estimates
what the call will cost, reserves that amount against the workspace, and once
the response is complete it settles the reservation against what was actually
spent. A request that would push the workspace past its cap is refused rather
than run and billed.

<Note>
  The reservation is an estimate, so a workspace close to its cap can see a
  request refused whose real cost would have fit. Raise the budget rather than
  retrying.
</Note>

Spend is measured over the calendar month and reads from the metrics recorded
for that workspace.

## When a quota is hit

The request is refused with `429` and a message naming the rule and its reset
time. Nothing is charged for a refused request.

Raise the limit, disable the rule or wait for the window to reset.