Using the LLM Input Node, inputs to be used in the LLM execution nodes can be predefined and entered, and automation flows can be built. However, an LLM input node can only be placed before an LLM execution node, and only one LLM input node can be used for an execution.

The LLM Input Node icon is located as shown above. Click to create a new LLM input node. There are two types of inputs: documents and snippets. We will first take a look at documents.

Documents #

1. Scope by document
   You can scope your search by selecting specific documents within the documents you have uploaded to the knowledge base. In other words, it is a method of specifying documents to be used as input through a search (optional) after specifying a range of documents. If you do not select a separate document, all uploaded documents are included as the target, and documents whose application status is set to OFF in the document tab are not included in the target even if they are selected in the node.
2. Hashtag Scope
   Specifying the search scope of documents through hashtags allows you to search only within the desired scope. If you do not set a hashtag, it will be searched without any restrictions. 
3. Allow users to search ON / OFF
4. Asking a question to the user
   If you set Allow users to search to ON, you can enter a message to be displayed when asking a user a question. 
5. Document search method
   If you set Allow users to search to ON, you can select the document search method.
   Within a document, you can select whether to perform a search for ① document contents (Contents) or ② search by titles (Title).
6. Save as the variable below
   The list of documents retrieved is saved in the specified variable. Variables can be created from the project settings menu or directly from the drop-down menu. 

Snippets #

A snippet is a short piece of data from multiple third-party services, such as messenger, mail, and calendar services. In this example, we will look at an integration with Google Calendar. To set up the integrations, please refer to this user guide for snippets and additional sources.

1. Asking a question to the user
   • You can enter a question message to display to the user.
   • Here, we created a message that encourages users to enter the schedule they want to search within Google Calendar.
2. Sources to search
   • You can specify that short text sources from third-party services such as Google Calendar, Slack, and Gmail can be searched. (However, there may be restrictions on the data size and pages to be imported.)
   • Google Calendar is set as the search source because it is a schedule search.
3. Save as variable below
   • Returns the result from a specific snippet and stores it in the specified variable. Variables can be created from the project settings menu or directly from the drop-down menu.
   • Store the input found by Google Calendar in a variable called SHI.

Connect with the Run LLM node as shown above. The prompt was quickly made utilizing the SHI variable. For more information on prompt management, see the user guide here.

Variables added as dynamic inputs on the prompt management page are suggested in the form of chips when editing prompts in the node.
*Here, the variable called SHI is added as a dynamic input.

What does it look like in action? Get information about your calendar for this example.