Skip to content
Nappai Documentation

Deep Research Agent

The Deep Research Agent is designed to help you find high-quality, thorough information from the web. Unlike a standard search that gives you a list of links, this agent acts like a dedicated researcher. It doesn’t just look once; it plans its search, gathers information, checks if it has enough data, and then digs deeper until it has a complete answer. This ensures that the information you get is not just a quick summary, but a well-researched insight.

How it Works

Think of this agent as a smart librarian or researcher working for you. Here is the simple process of how it handles your request:

  1. Understanding Your Question: It first analyzes the question you give it to make sure it understands exactly what you need. It might refine your question to get better results.
  2. Creating a Plan: Before searching, it creates a structured plan to cover the topic from different angles. This ensures it doesn’t miss important details.
  3. Searching the Web: It goes out to the web and finds relevant information based on its plan.
  4. Checking for Gaps: After finding information, it “thinks” about whether it has enough detail. If something is missing or unclear, it identifies that gap.
  5. Deepening the Research: If it finds gaps, it performs new, targeted searches to fill them. It repeats this “search and check” cycle until it is confident the information is complete and accurate.

It uses advanced AI techniques behind the scenes to manage this conversation with itself, ensuring the final result is reliable and thorough.

Connection & Credentials

This component does not require any external API keys or credentials to be configured by the user. It is ready to use immediately once added to your dashboard.

Inputs

Since this component inherits its configuration from a base system, specific input fields are managed by the underlying automation framework. In most cases, you will simply provide the topic or question you want researched. The system will handle the complexity of connecting this input to the agent’s internal planning logic.

Outputs

Once the research is complete, the agent provides a synthesized report. This output typically includes:

  • A comprehensive answer to your initial question.
  • Key findings and insights gathered from various web sources.
  • A summary of the research process, if available.

Output Data Example (JSON)

json { “final_answer”: “Based on our deep research, the primary causes of climate change in coastal regions include…”, “key_findings”: [ “Rising sea levels are accelerating due to thermal expansion.”, “Local government policies have reduced flood damage by 15%.” ], “sources_reviewed”: 12, “confidence_score”: 0.92 }

Connectivity

This component is typically placed in the middle of a workflow where data gathering is required.

  • Connected To (Input): It is usually connected to a Text Input component or a Data Extractor that provides the initial topic or query.
  • Connected To (Output): The output is often sent to a Text Formatter, a Summary Generator, or directly into a Document Creator to compile the research into a report or email.

Usage Example

Imagine you need to write a report on the current trends in renewable energy for your company.

  1. You create a Deep Research Agent in your Nappai dashboard.
  2. You connect a text box containing the question: “What are the latest trends in solar energy adoption in 2024?”
  3. The agent begins working. It doesn’t just list top 10 articles. It reads dozens of sources, compares different viewpoints, and checks for the most recent data.
  4. Once finished, it outputs a detailed summary that you can copy directly into your report.

Tips and Best Practices

  • Be Specific: While the agent is smart, providing a clear and specific question yields better results. Instead of “Tell me about cars,” try “Compare the fuel efficiency of hybrid SUVs in 2023.”
  • Allow Time for Processing: Since this agent performs multiple rounds of searching and reflection, it may take longer than a standard search component. Use this for important decisions where accuracy matters more than speed.
  • Iterative Refinement: If the first result isn’t quite what you needed, you can try refining your initial question and running the agent again rather than trying to edit the output manually.

Security Considerations

  • Data Privacy: This component interacts with the public web. Avoid using it to search for sensitive personal data or internal company secrets.
  • Source Verification: While the agent strives for accuracy, always review the final output before publishing, as it synthesizes information from external sources.