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

# Token usage

> Understand how tokens are consumed and manage your allowance.

## What are tokens?

Tokens are the unit of measurement for AI model usage. Every conversation your agent has consumes tokens:

* **Input tokens** — generated from the caller's speech (converted to text), plus your agent's instructions, knowledge base context, and tool responses
* **Output tokens** — the agent's spoken responses

## How tokens are counted

A typical call's token usage depends on:

| Factor | Impact on tokens |
| - | - |
| Call duration | Longer calls = more tokens |
| Instruction length | Longer instructions = more input tokens per turn |
| Knowledge base context | More retrieved chunks = more input tokens |
| Tool usage | Tool responses add input tokens |
| Response verbosity | Longer agent responses = more output tokens |

> [!TIP]
>
> Keep your instructions concise and focused. Verbose instructions consume extra input tokens on every single turn of every call.

## Viewing usage

Check your token usage in **Settings → Tokens**:

* **Current usage** — tokens consumed this billing period
* **Allowance** — your plan's monthly token limit
* **Usage percentage** — visual indicator of how much you've used

## Token top-ups

If you exceed your monthly token allowance, you can purchase top-ups:

**1. Go to token settings**

Navigate to **Settings → Tokens**.

**2. Select a top-up amount**

Choose from available token packages.

**3. Complete payment**

Pay via the secure payment page. Tokens are added immediately.

> [!NOTE]
>
> Top-up tokens don't expire at the end of the month. They're consumed after your monthly allowance is used up.

## Reducing token consumption

> [accordion] Shorten your instructions
>
> Remove redundant or overly detailed instructions. The AI model is good at inferring behavior from concise guidelines.

> [accordion] Optimize knowledge base
>
> Remove outdated or irrelevant documents. Fewer, higher-quality chunks mean less context per query.

> [accordion] Use a smaller model
>
> If your use case is straightforward, consider switching to `gpt-4o-mini-realtime` which uses fewer tokens per response.

> [accordion] Limit response length
>
> Set the **max output tokens** in your agent's model settings to cap response length.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.