Use Deep Research

Overview

Deep Research is a system that, with simple instructions/queries, can collect, analyze, synthesize, and infer various information from the web and documents to generate research reports for complex questions. Implemented via an LLM Agent, Alli's Deep Research feature makes it easy to create in-depth research reports using the web and Alli-linked documents as sources. Including the examples below, tasks that previously took hours to days can be executed in just minutes.

• When simple question-and-answer is insufficient and more diversified information gathering is needed

• When an in-depth research report including many cases is required

• When strategy formulation based on clear evidence is necessary

This guide explains how to use Deep Research in Alli and what settings are required to use Deep Research.

How to use Deep Research in Works

Use from the Works home screen

After accessing Works, activate Deep Research on the home screen as shown below. Enter your desired question and the Deep Research app will run. For detailed execution steps, refer to the "Using the Deep Research app" section below.

Using the Deep Research app

After accessing Works, enter the created Deep Research app. (How to create the app is described later.) Enter your desired question and the configured Deep Research Agent will propose investigation steps as shown below.

Proposal of investigation steps based on the user's question

After that, the Deep Research Agent asks additional questions needed to clarify the research requirements. Once the user answers, the research begins.

Confirm additional questions and start research

Research runs asynchronously and results are displayed as shown below when complete. It takes about 10 minutes to several tens of minutes, and you can check intermediate steps while it's in progress.

Result report

Also, before starting Deep Research, the user can specify the document search scope and web search scope. Choose from documents you have permission to access, or add trusted website URLs to limit the report generation to a restricted scope.

How to create a Deep Research app (dashboard access required)

When you create a new project through inquiry, a Deep Research app is embedded as the default app. In other cases, you can set it up as follows.

1. App Management > Create New App > Create Interactive App > Add Deep Research Node

   

   Add Deep Research node
2. Configure inside the Deep Research node

   1. You can choose how the question is entered. If you want to receive direct input from the user, select 'User Input' (member input); if you want to use a preconfigured variable as the question, select 'Variable'.

   2. The model and LLM agent are currently fixed to the highest-performing model and cannot be selected. They will be expanded in the future.

      

      Input settings within the node
   3. Sources to search: you can specify three options: documents, internet, and uploaded files (documents).

      1. Documents: If selected, the member/user running the app will include documents they have permission to access in the search scope. You can narrow the scope to specific documents by adjusting folders or hashtags.

      2. Internet: If selected, websites are included in the information search scope. If you add a specific site, only the websites that belong to that URL (and its subpaths) will be included in the search scope.

      3. Uploaded files: If selected, members/users can directly upload documents and include those documents in the information search scope. If not selected, uploading itself is not possible.

         

         Detailed scope settings within the node
3. After applying the above settings in the node, click 'Publish' to use the Deep Research app on Works/SDK.

   

   Use the app by publishing it

Other useful information

1. The Deep Research node can only be placed at the very end of the app. In other words, no further nodes can follow it.

2. Works users can make some additional adjustments to the node's 'Sources to search' settings.

   1. When setting the search scope in the node, if documents/internet/uploaded files are not selected, users in Works cannot arbitrarily use them. For example, if only documents are selected, searching the internet is not possible.

   2. If documents/internet/uploaded files are selected, users can specify ranges other than the node's predefined detailed scope within the permissions they have. That is, the values set in the node serve as initial values for the user's convenience,

3. When used in a chatbot embedded via SDK outside of Works, the user's ability to reset the scope is limited.