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

# Manage LLM Call Settings in Tyk AI Studio

> How to configure default call settings for Large Language Models (LLMs) in Tyk AI Studio, including parameters like temperature, max tokens, and more.

## Availability

| Edition | Deployment Type |
| :- | :- |
| [Community](https://tyk.io/docs/ai-management/ai-studio/overview.md#community-edition) & [Enterprise](https://tyk.io/docs/ai-management/ai-studio/overview.md#enterprise-edition) | Self-Managed |

Model call settings let you configure how Large Language Models handle prompts. These settings control parameters like response length, temperature (creativity level), and other options that shape the output when a prompt is sent to the LLM.

```mermaid theme={null}
graph LR
    A[User Prompt] --> B[AI Studio]
    B --> C{LLM Call Settings}
    C -->|Temperature, Max Tokens, etc.| D[LLM Provider]
    D --> E[Generated Response]
```

### Use cases

* **Chats**: These settings control how the LLM responds in conversational interfaces within the [Chat](https://tyk.io/docs/ai-management/ai-studio/chats.md) feature, allowing administrators to fine-tune the user experience.
* **Middleware Function Calls**: The settings guide LLM behavior in automated backend processes where the LLM is used for tasks such as data generation or content analysis.

## Managed List of Call Settings

A call setting is not a value you type inline for each chat. Each call setting is a saved entry in a managed list. You pick an existing entry when you [create a Chat](https://tyk.io/docs/ai-management/ai-studio/chats.md#configuration).

Tyk AI Studio seeds this list with default entries on first startup. The seeded list covers current models from major vendors, for example OpenAI, Anthropic, Google, and Meta. Seeded model names include `gpt-5`, `claude-opus-4-5-20251101`, `gemini-2.5-pro`, and `llama-3.3-70b`. This gives you a working set of models to use right away.

AI Studio seeds the list whenever it starts and finds the list empty - on a new installation, or if you delete every entry. Because seeding only happens when the list is empty, it never overwrites settings you have created or modified.

You can edit or delete any seeded entry. You can also add entries for other models. See [How to Create LLM Call Settings](#how-to-create-llm-call-settings).

## LLM Call Settings Details

The **LLM Call Settings** section allows administrators to configure default runtime parameters for Large Language Models (LLMs) used in chat interactions and middleware system function calls. These settings provide control over how the LLM processes inputs and generates outputs.

It is important to note that these settings are not utilized in the AI Gateway proxy (Tyk Edge Gateway). Applications created in the AI Portal by end users for accessing LLMs provide their own model configurations (like temperature and max tokens) in the API payload when making requests to the AI Gateway.

The Call Settings configured by the admin are specifically used for the built-in Chat Interface (accessed via the AI Portal) and for internal middleware system function calls (such as tool calling and RAG).

## Configuration

The **Edit/Create LLM Call Settings View** enables administrators to configure or update call-time options for a specific Large Language Model (LLM). Below is an explanation of each field and its purpose:

* **Model Name**: The name of the language model (e.g., `gpt-5.2`, `claude-opus-4-5-20251101`).
* **Temperature**: Controls randomness: `0` is deterministic, `1` is very random. Range: `0` to `1`.
* **Max Tokens**: The maximum number of tokens to generate in the response. Must be at least `1`.
* **Top P**: Controls diversity via nucleus sampling: `0.5` means half of all likelihood-weighted options are considered. Range: `0` to `1`.
* **Top K**: Controls diversity by limiting to `k` most likely tokens. `0` means no limit.
* **Min Length**: The minimum number of tokens to generate in the response.
* **Max Length**: The maximum number of overall tokens.
* **Repetition Penalty**: Penalizes repetition: `1.0` means no penalty, `>1.0` discourages repetition. Typically between `1.0` and `1.5`.
* **System Prompt**: A long-form text prompt that sets the context or behavior for the language model.

## How to Create LLM Call Settings

1. Navigate to **Chat > Model call settings** in the admin UI sidebar.
2. Click **Add call setting**.
3. Fill in the required fields such as **Model Name** and configure the desired parameters like **Temperature** and **Max Tokens**.
4. Click **Add call setting** to save the configuration.

   ![Create LLM Call Settings UI](https://mintcdn.com/tyk/hBFyexPCuAxneHvv/img/ai-management/ai-portal-llm-call-setting.png?fit=max\&auto=format\&n=hBFyexPCuAxneHvv\&q=85\&s=8a1ca8e42895c4ed56eb2bc739587524)

Once saved, the new entry appears in the managed list. You select it from this list when you [create a Chat](https://tyk.io/docs/ai-management/ai-studio/chats.md#configuration).