# Settings

{% hint style="success" %}
**Change Settings anytime:**\
\&#xNAN;*LLM Vision Settings → ... → Reconfigure*  
{% endhint %}

## Fallback Provider

If a request to your selected AI provider fails, LLM Vision can automatically retry with another provider. This is helpful if you get rate-limited or run out of credits.

<figure><img src="https://2802862115-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FFIhUccwfwWnHypSNsrKL%2Fuploads%2FDhy6LUr3eUW6NCNCYjJc%2Fimage.png?alt=media&#x26;token=61647f8d-91d1-4183-8319-65ca4a3d48f2" alt="" width="563"><figcaption></figcaption></figure>

## System Prompts

System prompts are special instructions that guide how the AI behaves. Unlike regular messages, system prompts act like internal rules or roles.

You can customize the system prompt for AI descriptions as well as title responses. Titles are used to generate event titles for the timeline and [blueprint](https://llmvision.gitbook.io/getting-started/setup/blueprint) notifications.

## Timeline

<figure><img src="https://2802862115-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FFIhUccwfwWnHypSNsrKL%2Fuploads%2FMZHF0ZfzW2dUMd0jtllC%2Ftimeline.png?alt=media&#x26;token=1ae13c16-7a5d-47c6-95fe-248bc89f2ee9" alt="" width="375"><figcaption><p>LLM Vision Timeline Card</p></figcaption></figure>

LLM Vision can remember analyzed events in a timeline you can view on your dashboard using the [Timeline Card](https://llmvision.gitbook.io/getting-started/setup/timeline-card).

## Memory (Beta)

<figure><img src="https://2802862115-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FFIhUccwfwWnHypSNsrKL%2Fuploads%2FAO3ZbLHnev4zZZZ7ozAs%2Fimage.png?alt=media&#x26;token=90656cc5-adad-48c4-b229-a9f674ba389e" alt="" width="563"><figcaption></figcaption></figure>

LLM Vision includes a built-in memory to store reference images along with descriptions to provide additional context for models. This is useful for [recognizing people, pets, cars, and other familiar objects](#user-content-fn-1)[^1].

To store an image in memory you need to provide a file path to the image and an image description. If you have multiple images, the order of images must be the same as the order of image image descriptions.

To use information stored in memory, enable `use_memory`which is available in `image`, `video`, `stream` and `data` analyzer.

[^1]: LLM Vision uses large language models. Do not use LLM Vision to automatically unlock doors or for similarly critical actions!&#x20;