> ## Documentation Index
> Fetch the complete documentation index at: https://docs.varios-ai.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Chat Models
> Configuration of chat models for text generation and conversations
Chat models are the core component for AI-powered conversations in VARIOS AI. They support text generation, document processing and optionally code execution.
## Basic Data
| Field | Required | Description |
| --------- | -------- | -------------------------------------------------------------- |
| **Image** | No | Profile image for the model (displayed in the model selection) |
| **Title** | Yes | Display name of the model (e.g. "Azure GPT 5.2") |
## API Configuration
| Field | Required | Description |
| --------------- | -------- | --------------------------------------------------------------------------- |
| **API** | Yes | API type – see [Available API Types](#available-api-types) below. |
| **Credentials** | Yes | Stored credentials for the selected provider (dropdown selection). |
| **Model Name** | Yes | Technical model name sent to the API (e.g. `gpt-5.2`, `claude-sonnet-4-5`). |
### Available API Types
| API Type | Description |
| ----------------------------------------- | -------------------------------------------------------- |
| **OpenAI Chat API** (`/chat/completions`) | Standard OpenAI Chat Completions API for text generation |
| **OpenAI Assistant API** (`/assistants`) | OpenAI Assistant API for structured assistants |
| **OpenAI Responses API** (`/responses`) | OpenAI Responses API for specific response formats |
| **IONOS API** (`/chat/completions`) | IONOS-compatible Chat Completions API |
| **OpenAI Chat API without HTTP Proxy** | Direct connection without proxy support |
## Costs and Limits
| Field | Required | Description |
| ---------------------------------------- | -------- | ------------------------------------------------------------------------------------------ |
| **Cost in \$ per million input tokens** | No | Cost per one million input tokens above the context threshold. |
| **Cost in \$ per million output tokens** | No | Cost per one million output tokens above the context threshold. |
| **Cost in \$ per million cache tokens** | No | Cost per one million cache tokens above the context threshold. |
| **Context Window** | No | Maximum number of tokens that can be processed in the context of a request (e.g. 1050000). |
| **Output Token Limit** | No | Maximum number of tokens that a model may generate in a response (e.g. 128000). |
If you want to avoid the higher costs of a pricing tier, choose a context window that is less than or equal to the configured threshold.
## Behavior Settings
| Field | Required | Description |
| ------------------------- | -------- | ------------------------------------------------------------------------------------------------------------------------------------ |
| **Temperature** | No | Controls creativity (0–2). Checkbox **Use temperature** must be enabled. Low = deterministic, high = creative. |
| **Top P** | No | Nucleus sampling parameter (0–1). Checkbox **Use Top P** must be enabled. Alternative to temperature control. |
| **Additional Parameters** | No | Additional API parameters as a JSON object. Enter valid JSON (e.g. `{}`). These parameters are appended directly to the API request. |
| **Thinking Effort** | No | Cognitive effort of the model. Checkbox **Use thinking effort** must be enabled. Value: **Low**, **Medium** or **High**. |
Allow the model behavior settings to be overridden in order to tailor your assistants even more precisely to your needs. (Expert view required)
## Streaming
| Field | Required | Description |
| ---------------------- | -------- | ----------------------------------------------------------------------------------------------------- |
| **Supports Streaming** | No | Checkbox **Use a streaming model** – Enables streaming responses (answer is displayed progressively). |
| **Stream Options** | No | Checkbox **Use stream options** – Enables extended streaming options (e.g. token usage in stream). |
## User Identification and MIME Types
| Field | Required | Description |
| ----------------------- | -------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **User Identification** | No | Checkbox **Use user identification** – Adds a unique user identifier to the API request. |
| **Allowed MIME Types** | No | File types that are sent **directly to the model** (e.g. images for vision models). All other file types are automatically handled via document processing. |
The **Allowed MIME Types** determine which files go directly to the model
without being routed through document processing. Typical example:
`image/png`, `image/jpeg` for vision models.
## Check Service Status
* Define whether test tokens should be sent to check the model's availability and at what interval this should happen. Costs are incurred for each check.
## DLP Security Settings
### Basic DLP Settings
#### GDPR Protection
Detects personal data in the chat and encrypts it before transfer to the AI.
| Option | Description |
| ------------ | ------------------------------------------------------------------ |
| **enabled** | GDPR protection is always active and cannot be disabled by users. |
| **optional** | GDPR protection is active by default but can be disabled by users. |
| **disabled** | GDPR protection is disabled. |
#### Sensitive File Protection
Blocks the upload of documents with confidential classification.
| Option | Description |
| ------------ | -------------------------------------------- |
| **enabled** | Upload of confidential documents is blocked. |
| **disabled** | Upload of confidential documents is allowed. |
### Enterprise DLP Settings
#### DLP Interface (ICAP)
Sends texts and files for review to the configured ICAP server.
| Option | Description |
| ------------ | ------------------------------------ |
| **enabled** | ICAP server is used for all content. |
| **disabled** | ICAP server is not used. |