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

# LLM Settings

> Choose providers, models, and temperature settings

The LLM (Large Language Model) powers your agent's conversation intelligence — understanding callers and generating responses.

## Providers and models

| Provider   | Model                        | Type            | Notes                             |
| ---------- | ---------------------------- | --------------- | --------------------------------- |
| **OpenAI** | GPT 4.1                      | Standard        | Best overall quality              |
| **Azure**  | GPT 4.1                      | Standard        | Same model, hosted on Azure       |
| **OpenAI** | GPT 4.1 mini                 | Standard        | Faster, lower cost                |
| **Google** | Gemini 2.5 Flash             | Standard        | Fast with good quality            |
| **Google** | Gemini 2.5 Fast Live         | Native Realtime | Ultra-low latency, built-in voice |
| **Google** | Gemini 2.5 Fast Live Cascade | Cascade         | Realtime with text fallback       |
| **OpenAI** | Realtime                     | Native Realtime | Ultra-low latency, built-in voice |
| **OpenAI** | Realtime Cascade             | Cascade         | Realtime with text fallback       |

### Standard vs Realtime vs Cascade

* **Standard** — Text-based LLM with separate TTS/STT. Most flexible voice options. Best for complex conversations.
* **Native Realtime** — Voice-native model with built-in speech. Lowest latency. Limited voice selection (provider's built-in voices only).
* **Cascade** — Starts with realtime for fast initial response, then falls back to text-based processing. Good balance of speed and quality.

## Temperature

Controls how creative vs deterministic the agent's responses are.

| Value         | Behavior                          | Best for                                     |
| ------------- | --------------------------------- | -------------------------------------------- |
| **0.0 - 0.3** | Very consistent, predictable      | Factual Q\&A, compliance-sensitive scenarios |
| **0.4 - 0.7** | Balanced (default: 0.7)           | General customer service                     |
| **0.8 - 1.2** | More varied, creative             | Casual conversations                         |
| **1.3 - 2.0** | Highly creative, less predictable | Not recommended for production               |

## Impact on other settings

When you select a **Native Realtime** provider:

* **TTS provider** is automatically set to the realtime provider's built-in voice
* **STT provider** is set to null (built-in)
* **Voice selection** switches to the provider's available realtime voices
* **Speaking rate** is not adjustable

## Who can edit

Only **Super-Admin** and **Dev-Admin** roles can configure LLM settings. Client-Admin and Client-Employee users don't see this section.