> For the complete documentation index, see [llms.txt](https://docs.xenioo.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.xenioo.com/ai/large-language-models/settings.md).

# Settings

To integrate the knowledge and capabilities of an LLM into your chatbot, you need to configure some basic settings.<br>

<figure><img src="/files/JAsirwDasSXslKS7Cvpa" alt=""><figcaption></figcaption></figure>

### Service Provider

Choose one of the available service providers, such as OpenAI or Gemini.

Ensure you have set up an account with your chosen service provider to access their language model services.

### Model Version

When selecting the desired model version, please review the available options carefully. Consider the features and specifications of each version to ensure it meets your requirements. It's important to evaluate performance metrics, compatibility, costs and any other relevant criteria before making a choice.

{% hint style="warning" %}
We currently support a limited number of language models. Please note that while we might introduce support for additional models in the future, older models may also lose support over time.
{% endhint %}

### Integration API Key

The integration API Key let you attach you provider API to Xenioo. Depending on the provider you've selected you might need to get this API key differently. [OpenAI ](https://help.openai.com/en/articles/4936850-where-do-i-find-my-openai-api-key)will require you to generate a new api key from the account section while with [Google Gemini ](https://ai.google.dev/gemini-api/docs/api-key)you will need the account access full JSON.

By default, the API Key will be hidden. Click on show to display the full value.

{% hint style="warning" %}
The API Key text is obscured for security reasons. Each time an user clicks the "SHOW" button, this action is recorded in the Audit Logs.
{% endhint %}

### Advanced Settings

<figure><img src="/files/qvj8Vhsz2R3Y1cgyUo8q" alt=""><figcaption></figcaption></figure>

### Caching

Caching lets Xenioo reuse the repeated part of the request sent to the AI provider.

This usually includes static content that stays the same across multiple calls, such as guidelines and other repeated context.

This can reduce input processing time and lower cost on long prompts.

The model still generates a new answer each time. It does not replay an old reply.

Xenioo handles the provider-specific caching method for you. You do not need to build a custom request or configure provider-specific cache rules manually.

### Cache modes

Different providers use different cache modes.

**OpenAI**

* `5 mins / 1 hour (automatic)` keeps the cached prefix available for short-term reuse during active usage.
* `24 hours` keeps the cached prefix available for reuse for up to 24 hours.

**Google Gemini**

* `explicit` lets Xenioo create and reuse cached content with a custom TTL.

Implicit caching can still happen automatically on the provider side, but it is not a selectable cache mode in Xenioo.

**Anthropic**

* `5 minutes` keeps the cached prefix available for up to 5 minutes.
* `1 hour` keeps the cached prefix available for up to 1 hour.

### Models that support caching

**OpenAI**

These models support caching in Xenioo:

* `gpt-5`
* `gpt-4o`
* `gpt-4o-2024-11-20`
* `gpt-4o-mini`
* `gpt-4o-mini-2024-07-18`

These models do not support it:

* `gpt-4-turbo`
* `gpt-4`
* `gpt-3.5-turbo`

**Google Gemini**

These models support caching in Xenioo:

* `gemini-2.5-pro`
* `gemini-2.5-flash`

This model does not support caching:

* `gemini-2.5-flash-lite`

Gemini caching in Xenioo uses explicit cached content with a custom TTL.

**Anthropic**

These models support caching in Xenioo:

* `claude-opus-4-7`
* `claude-sonnet-4-6`
* `claude-haiku-4-5`

Anthropic uses its own caching method behind the scenes. Xenioo manages it for you, so the setting still works at the product level without any custom request setup.

### Minimum tokens required to trigger caching

Providers usually require a minimum amount of repeated input before caching can apply.

**OpenAI**

Caching starts at **1,024 prompt tokens**.

After that, the cached prefix grows in **128-token increments**.

**Google Gemini**

Gemini uses explicit caching in Xenioo.

Minimum cached input size depends on the model:

* `gemini-2.5-flash` — **1,024 tokens**
* `gemini-2.5-pro` — **4,096 tokens**

**Anthropic**Caching starts when the cacheable prefix reaches at least **1,024 tokens**.This minimum applies to all supported Anthropic models in Xenioo:

* `claude-opus-4-7` — **2,048 tokens**
* `claude-sonnet-4-6` — **2,048 tokens**
* `claude-haiku-4-5` — **4,096 tokens**

This threshold applies to the cached prefix before the provider reuses it.