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

# Limits and costs

> How credits are spent, how to estimate a job before running it, and what stops a run from getting out of hand.

## How credits are spent

Credits are the currency behind everything the assistant does. 1 credit equals \$0.01. A plain
conversation with the assistant, with no device work, doesn't burn meaningful credits.

Credits start moving once the assistant acts, on average around **0.5 credits per step**. A step
is one unit of work the assistant takes toward your goal, not one tap on the screen. A single step
can bundle several taps, screenshots, or waits together, so tightly described tasks cost less than
vague, exploratory ones.

<Note>
  Your device subscription includes a monthly credit allowance that refreshes each billing cycle.
  Top-up credits stay until you spend them. See [Credits](/credits) for the exact allowance per
  device type.
</Note>

## Getting a cost estimate

You can ask the assistant how much a task will cost before you run it. It gives you a rough
estimate based on how many steps the task is likely to take, not a hard quote. The only exact
number is what a run actually spends, which you can see on the task afterward or in your account
balance.

> Roughly how many credits would it cost to scrape this page every day?

<Tip>
  You can also ask the assistant to explain how credits are counted, or to show you a breakdown of
  your recent usage, straight in the chat.
</Tip>

## What makes a job expensive

* **More steps.** A task that decides what to do screen by screen, using vision to figure out
  where to tap, takes more steps than one with a clear, repeatable path.
* **Loosely described goals.** "Check my accounts" takes more exploration than "open Instagram and
  check my DMs."
* **Extra devices.** One Cloud Phone can usually run several accounts. Add a device only when you
  genuinely need separate isolation, such as a dedicated proxy per account.
* **Repeated retries.** If the same approach keeps failing, the assistant treats that as a signal
  to change tactics or stop, rather than a reason to try the exact same thing again.

## Budgets that stop runaway jobs

Every task has a cap on how many steps it can take and how long it can run, so a stuck task can't
burn credits indefinitely. While it works, the assistant also checks whether it's actually making
progress toward your goal. If a run stops advancing, it stops itself, reports what it achieved, and
tells you what's left, instead of looping.

A Pay as you go device also stops billing on its own: it shuts down automatically if your credit
balance runs out.

## What it refuses or hands to the dashboard

The assistant never buys, subscribes, upgrades, or adds credits on your behalf. Anything involving
money shows up as a card in the chat, and you complete it in the dashboard yourself. It does not quote
plan prices from memory; it shows you the current pricing card. For your balance, check the
Billing section of the dashboard.

## Plans, upgrades, and top-ups

Ask the assistant what plans are available or how to add credits, and it shows you the relevant
card. See [Credits](/credits) for how device subscriptions, credit allowances, and top-ups work.

## Related

<CardGroup cols={2}>
  <Card title="Credits" icon="coins" href="/credits">
    The full credit system, device allowances, and top-up pricing.
  </Card>

  <Card title="Overview" icon="compass" href="/assistant/overview">
    What the assistant is and what it can do.
  </Card>

  <Card title="Human in the loop" icon="user-check" href="/assistant/human-in-the-loop">
    How approvals and questions pause a run mid-task.
  </Card>

  <Card title="Getting started" icon="rocket" href="/assistant/getting-started">
    Your first conversation and first result.
  </Card>
</CardGroup>
