When you connect a datasource to your app with Intelligent Search turned off, elvex will always search it using the exact text of the last message you sent. If you have multiple datasources connected, elvex will search against all of these datasources and return the top 5 results across them all.

When you enable Intelligent Search, this default search behavior is disabled. Instead the app will determine if a datasource search is needed, what the search term will be (which can differ from the message) and which datasources to search (if at all). Your app's rules and instructions as well as your datasource names and descriptions now guide what is searched and when.

With the app in control of what to search for and when it can improve its search based on your follow up questions and the context of the conversation. It can also selectively choose to search different datasources based on the conversational context.
​

This setting should only be used for apps that do not require grounding for every response. If an app is not grounded, this means you are allowing the app to use knowledge from sources other than the attached datasources to respond to your request. Other sources would include baseline knowledge from the LLM's training dataset and information the app finds from searching the internet (if this feature is turned on for the app you are using).

Apps that are not grounded can be more likely to hallucinate (provide a response that is not factual), so it's important to think through the app's use case carefully before turning this feature on.

Good times to use it:

​

Apps with this feature enabled will not always search your connected datasources for material to ground their answers. The app will decide when to search and what to search. This means that your app rules, context, and datasource descriptions become incredibly important.

If you're not sure whether you should turn this feature on for your apps, please email us at [email protected] and we'll be happy to help!