Agent Builder

Updated on June 22, 2025

The Agent Builder menu allows you to easily build AI agents that can be used in your business. This guide introduces a series of steps to build an AI agent using single-action apps in the Alli platform and the MCP (Multi-Component Platform) server and tools provided by our company.

Agents can automate various business workflows and execute processes by linking with multiple tools. As an example, we will introduce an example of building an agent that can perform web searches and article summaries using apps created with Alli.

How to create an agent #

1. Go to the Agent menu in your Alli dashboard.

2. Click the “+Create Agent” button.

3. Enter the agent name, icon, and a brief description, then click the Save button.

Select the model to use with the agent and enter instructions #

1. Select the model you want your agent to use. 2. Enter instructions that specifically describe the task your agent will perform.

Instruction Example #

# Role
You are a "Web Search + News Article Summary Agent."

# Purpose
Perform web searches and summarize news articles in response to user requests.

Instructions, like prompts, are important for LLMs to understand their role correctly. When configuring more advanced agents, it is recommended to clearly describe the purpose, the specific actions the agent should take, and when to call the tools (described below).

Adding Tools #

Set the tools that the agent can execute. This time, we will use the following two tools.

News article summary app with translation function

1. This time, we will use an app from the App Market that can be installed immediately. Add the “News Article Summary: With Translation” app from the “App Market” menu.

2. In the “Agent” menu, open the agent’s detailed settings page and click “+ Add” in the Tools column.

3. From the tool selection screen, select “Allganize Alli” and select the app you added in step 1. *As of May 22, 2025, only Single Action type apps can be selected as agent tools.

4. If your tool requires inputs, enter values. This is only to be set if you need your agent to use that value for its output. The inputs that can be set within a tool are based on the concept of “overrides”. If you don’t need to use a specific value, you can leave it blank.

In this app, there is no need to set specific values ​​for the input items, so leave the following settings blank.

If you enter a specific value, there are no restrictions on the format of the string you enter into the input – what you enter will be automatically added and used as part of the agent’s instructions.

This time we used the default app from the app market, but you can also specify your own customized app as the tool. We encourage you to create an app that fits your company’s business workflow and enhance the agent’s functions.

Bing: bing_web_search

Next, add the bing_web_search tool. There are various tools in Bing’s MCP server, but this time we will add bing_web_search to perform a simple web search.

1.From the tool selection screen, select “Bing (Allganize)” and then select “bing_web_search”.

2. Bing (Allganize) MCP server and bing_web_search tool have been added.

①Bing (Allganize) MCP server settings

Set the API key and endpoint required for the Bing API. Please specify the Bing API that your company has contracted with. When the agent calls each tool registered on the Bing (Allganize) MCP server, the authentication information will be overridden with the contents set here.

②Bing_web_search tool settings

If you need to specify a specific value, please enter it in each setting item. Please refer to the following for the input items of this tool.

  • query: Normally, when a user asks a question, the agent analyzes it and automatically performs multiple web searches. However, if you want to use a specific query, enter the query in this field. If you don’t need to specify anything, you can leave it blank.
  • count: The agent may autonomously adjust the total number of search results depending on the situation, such as limiting the search results to a maximum of 5 due to model token constraints. However, if you want to make sure that a web search always returns 100 results, you can set the count to 100 to retrieve the specified number of results. If you do not need to specify anything, you can leave this field blank.
  • offset: For example, if you don’t need to check all the search results retrieved by the agent from the beginning, but want to start the search from the top 10 search results, you can specify this in the setting item. If you don’t need to specify anything in particular, you can leave this field blank.
  • market: This is the item to set the language of the website to be searched. For example, if you want to search English websites, you can specify EN. If you do not need to specify a language, you can leave it blank.

The tool setup is now complete.

Release and implementation of the agent #

1. Once you have created your agent, publish it by clicking the “Publish” button in the top right.

2. Create a new conversational app and implement the agent you just created. Create a flow like the one below.

① Just place the “Conversation response node” as shown below without setting any optional items.

②Please set the “LLM execution node” as follows.

(1) Specify “Agent” as the execution type.

(2) For “Agent”, select the agent you just created from the list.

(3) For “Base model”, be sure to specify the same model as the agent.

Using the Agent #

Let’s try using the implemented agent on Alli Works. Enter a query like the following:

The agent then uses the instructions to carry out the necessary operations.

Finally, a summary was output based on the agent’s own web search results and following the prompts of the app that was added as a tool.

In addition to the MCP tools introduced here, we have many other tools available and plan to add more in the future.

Table of contents