How to set up and use Research Agent action
Searches and navigates the internet to find information.
The Research Agent is a versatile tool that empowers you to leverage AI for complex research tasks. Unlike more predefined actions, the Research Agent can autonomously navigate the internet, find and analyze information, and adapt to deliver the results you need.
With the Research Agent, you have the flexibility to guide its actions by providing instructions. This allows you to handle a wide range of research scenarios that might be harder to address with predefined actions alone. Whether you need to gather data for competitor analysis, generate leads, or conduct market research, the Research Agent can intelligently search the web, extract relevant information, and compile it for you.
Compared to actions like "Summarize LinkedIn Profile," which requires a specific input (the profile URL), the Research Agent can work with the data you provide and find additional information as needed to complete the task at hand. This adaptability makes it a powerful tool for handling complex research tasks that require creative problem-solving and logical reasoning, much like a human researcher would.
To set up a research agent, follow the steps below:
There are two ways to use the Research Agent action in Cassidy: pre-built agents and custom agents.
Pre-built Research Agents come with predefined instructions and are designed for common use cases, making them easy to set up and use.
Custom Research Agents allow you to provide your own instructions, giving you more flexibility and control over the agent's behavior.
In this guide, we'll explore both options. For the pre-built example, we'll use the "Find LinkedIn Company Profile" agent, and for the custom example, we'll create an agent to verify domain status of a website.
Pre-built Research Agents:
Add a Research Agent action: While editing a workflow, click the plus button between existing blocks. This will open a modal where you can select from Cassidy's pre-configured actions. Choose the Research Agent action to add it to your workflow. *We'll use the "Find LinkedIn Company Profile" agent as an example for this documentation.
Rename the action (if needed): If you want to make the action more descriptive or easier to reference as a variable later, enter a new name.
Input required fields: Enter the information required by the specific pre-built agent. For the LinkedIn Company Profile example, input the Company Name and Company Website.
*The agent instructions are pre-populated and will automatically use the inputs you provided. To see what steps it will take, open the Advanced Options tab and view the "Instructions for the Research Agent" field. You will see the variable references populated in the instructions.
If you want to modify the default instructions, click "Convert to custom Research Agent (Advanced)". This will create a custom research agent that you can edit. Refer to the custom agent setup steps below for tips on writing effective instructions.
Review the output: When running the workflow, the Research Agent output will display the information you requested. You can also open the Research Agent Details to see the confidence level, reasoning behind the result, and the steps the agent took to obtain the information.
Custom Research Agents:
Add a Research Agent action: While editing a workflow, click the plus button and select "Research Agent" from the actions list. For this example, we'll create a custom agent to verify domain status.
Rename the action (if needed): If you want to make the action more descriptive or easier to reference as a variable later, enter a new name.
Enter "Instructions for the Agent": Provide detailed instructions describing the task you want the Agent to perform.
Be specific about the desired response and potential information sources.
The Agent can search the internet, scrape webpages, and perform multiple steps autonomously.
Reference variables to provide necessary context.
Include directions and the order of operations.
Specify when assumptions can be made.
Clarify what the Agent shouldn't do.
Define the expected output format (e.g., "valid domain" for a positive case or "no results found" for a negative case).
Review the output: When running the workflow with the custom Research Agent, review the output to ensure it meets your expectations. Check the Research Agent Details for confidence level, reasoning, and the steps taken. Refine the instructions if needed for better results.
