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

# AI Assistant credits & usage

> How AI Assistant usage is measured in credits, the monthly allowance included with your contract, what happens when it runs out, and what affects how quickly credits are used.

Use of the AI Assistant is measured in **credits**. This page explains where your credits come from, what happens when they run out, and what affects how quickly they're used — so you can plan usage across your team.

## Your credit allowance

Your AI credits are a **monthly allowance shared across your subdomain**. Everyone on your subdomain draws from the same pool — questions asked from the interface and actions triggered by automations both count against it — and the allowance **refreshes at the start of each calendar month**.

Every subdomain includes a **default allowance**. Your contract can set a **higher recurring allowance**, and your allowance can be **topped up** if you need more within a month.

<Info>
  Your allowance is managed by Clarus. To move to a higher monthly tier or add more credits, talk to your account manager.
</Info>

## What happens when credits run out

If your subdomain uses up its monthly allowance, the assistant **pauses new requests** until the allowance refreshes at the start of the next month, or until it's increased. This includes actions triggered by automations.

<Note>
  The check happens **before** any request is sent to a model, so you can't accidentally run over your allowance — the assistant simply stops until there are credits available again.
</Note>

## What uses credits

Credits reflect how much work each request involves — specifically the amount of text the assistant has to read in and generate back. The main factors:

* **Which model you choose** — this is the biggest lever. More capable models (such as Claude Opus) use considerably more per request than lighter, faster ones (such as Claude Haiku). The model picker shows a **cost indicator** for each option so you can choose with that in mind.
* **How much data a question needs** — questions that make the assistant retrieve and read a lot of records use more than a quick, narrow lookup.
* **The length and complexity of a conversation** — longer threads carry more context, so a long back-and-forth uses more than a single focused question.
* **How much you use it** — the total number of conversations and actions across your team.
* **Automations** — unattended AI actions draw on the same monthly allowance as people do, and can add up if they run frequently.

<Tip>
  You can see usage as you go: the **model picker** shows each model's relative cost, and each answer includes a short footer with the tokens it used. That makes it easy to spot which kinds of question are heavier.
</Tip>

## Getting the most from your allowance

* **Match the model to the task** — use a lighter model (such as Claude Haiku) for routine, everyday questions, and reserve the most capable model for genuinely complex analysis. See [Overview](/knowledgebase/ai-assistant/overview) for the options.
* **Ask focused questions** rather than open-ended ones that pull in more data than you need.
* **Scope automations** to what they actually need, so scheduled or high-volume runs don't consume more than intended. See [Access & controls](/knowledgebase/ai-assistant/access-and-controls).