# Plans and usage

> What consumes your allowance, how the weekly limit works, and where to see how much you have used.

Endue meters the thing that actually costs money: **model usage**. Everything else — agents, conversations, artifacts, connectors — is not billed per action.

## What consumes usage

Every [run](/docs/work/runs/) sends your conversation, the agent's prompt, and its tool results to a model, and the model answers. Both directions count.

That means usage is driven by:

| Driver | Effect |
| --- | --- |
| **Which model** | The largest factor by far — prices across the catalog differ by more than 10× |
| **[Reasoning effort](/docs/build/models/#reasoning-effort)** | Higher effort spends more tokens before answering |
| **How many steps a run takes** | Each tool result goes back into the model |
| **Conversation length** | A long thread is re-read on every step |
| **[Routines](/docs/automate/routines/)** | They run whether or not you are watching, on their own schedule |

## The weekly limit

Your plan carries an allowance measured over a **rolling week**, not a monthly balance you can spend all at once. When the week's allowance is used up, runs are declined until the window moves on.

Two consequences worth planning around:

- A single heavy day cannot exhaust a month.
- A routine on a short interval with an expensive model is the most common way to hit the limit, because it spends steadily while you are not watching.

<Aside type="note" title="Where to see the current numbers">
  Your plan, its allowance, and what you have used are in **Settings → Usage**.
  Current prices and what each plan includes are on the pricing page in the app —
  those change, and the app is the authoritative place to read them.
</Aside>

## Keeping usage down

In rough order of effect:

1. **Match the model to the job.** A fast, cheap model for routine work, a strong one for the few hard messages. See [Choosing a model](/docs/build/models/).
2. **Pin a model on each routine**, so scheduled work has a predictable cost.
3. **Lower reasoning effort** for work that is not multi-step.
4. **Start new conversations** when the topic changes — long threads are re-sent on every step.
5. **Turn off [built-in tool groups](/docs/capabilities/built-in-tools/)** an agent does not need; every advertised tool is context on every step.
6. **Widen routine intervals.** Hourly is rarely worth four times what four-hourly costs.

## Bring your own key

If you would rather pay a model provider directly, connect your own key and Endue will use it for that model's calls. See [Bring your own key](/docs/account/bring-your-own-key/).

## Limits

- Usage is metered per account, not per agent. The usage view shows the account total.
- Runs are declined when the weekly allowance is exhausted. Work in flight is not rolled back, but new runs will not start.
- Figures in the usage view can lag a run by a short period while it is settled.

## Related

<CardGrid>
  <LinkCard
    title="Choosing a model"
    href="/docs/build/models/"
    description="The single biggest lever on what you spend."
  />
  <LinkCard
    title="Bring your own key"
    href="/docs/account/bring-your-own-key/"
    description="Pay the provider directly instead."
  />
  <LinkCard
    title="Routines"
    href="/docs/automate/routines/"
    description="Scheduled work — the usage you are least likely to notice."
  />
</CardGrid>