> ## 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.

# AI models

> Choose the AI model that powers your agent's conversations.

## Available models

AssistPulse uses OpenAI's Realtime API for voice conversations. The model determines the quality and capability of your agent's responses.

| Model | Best for |
| - | - |
| **gpt-4o-realtime** | Most conversations — excellent balance of speed and intelligence |
| **gpt-4o-mini-realtime** | High-volume, simpler use cases — faster and cheaper |

> [!NOTE]
>
> Model availability may change as new models are released. The default model is recommended for most use cases.

## Model settings

### Temperature

Controls how creative or deterministic the agent's responses are:

* **0.0** — highly deterministic, always picks the most likely response
* **0.5** — balanced (default)
* **1.0** — more creative and varied responses

> [!TIP]
>
> For customer service and reception agents, keep the temperature between 0.3 and 0.7. Higher values can lead to inconsistent or unexpected answers.

### Max output tokens

Limits the length of each response. Options:

* **Default** — no explicit limit (model decides)
* **Custom value** — set a specific token limit

> [!WARNING]
>
> Setting the max output tokens too low may cause the agent to cut off mid-sentence. Only adjust this if you need to control response length for cost reasons.

### Tool choice

Controls how the agent decides when to use connected tools:

* **Auto** — the model decides when a tool is relevant (recommended)
* **Required** — forces the model to use a tool on every turn
* **None** — disables tool usage entirely

## Token consumption

Every conversation consumes tokens:

* **Input tokens** — the caller's speech converted to text, plus instructions and knowledge context
* **Output tokens** — the agent's spoken responses

Token usage is tracked per call and visible in your call history. See [token usage](https://docs.assist.fm/billing/tokens) for details on limits and top-ups.


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