AI Powered Search Agent

Agent Description:

The Search Agent is a dual-source intelligent assistant that answers user questions by consulting both internal knowledge bases and live web content. It ensures users receive accurate, relevant, and up-to-date responses — whether the answer resides in company documents or must be sourced externally.

Purpose and Components
  • Purpose:

    To provide users with comprehensive answers to questions by querying an internal knowledge base and performing live web searches when internal answers are unavailable or outdated.

  • Components:
    • A knowledge retrieval agent to semantically query internal FAQs and policy files.
    • A response agent that summarizes internal results or triggers web search fallback logic.
Supported Capabilities
  • Interpreting natural language user queries.
  • Performing semantic searches on internal documents (e.g., product FAQs, help docs).
  • Automatically falling back to web search when internal sources are insufficient.
  • Synthesizing concise, context-aware responses using multi-source information.
  • Providing source attributes for web-sourced answers.
LLM Used
  • Google Vertex
    Note: To learn more about the LLM and to modify its behavior, refer to the Configuring LLM settings section.

Sub-Agents

1. Knowledge Retrieval Agent

  • Role: Internal knowledge base search engine using RAG.
  • Scope: Perform RAG-based search on company documents to retrieve answers.
  • Description: This subagent accepts a user query (e.g., “How do I reset my password?”) and performs a semantic search over the internal knowledge base (e.g., product_faqs.txt) using the RAG technique. If the query yields a complete and relevant result from internal content, it passes the result to the Search Agent for summarization. If the retrieved information is outdated or missing, it instructs the Search Agent to conduct a web search instead.
  • Tools used: MCP Client Tool
  • LLM Used: Google Vertex (inherits from parent).

2. Search Agent

  • Role:Generates final answers from internal or external sources.
  • Scope:Summarize internal data or search the web for additional answers.
  • Description: This subagent receives either:

    • Retrieved knowledge base content for summarization (Scenario A), or

    • A fallback instruction to perform a web search (Scenario B).

    • In Scenario A, it summarizes the relevant internal answer into a clear user-friendly response.

    • In Scenario B, it searches the web using the query, analyzes top results, and composes a concise answer, prefixed with:

      • This answer was sourced from the web and is not from our official FAQ

      • If no valid answer is found online either, it gracefully responds with: I could not find a reliable answer in our knowledge base or on the web.
  • LLM Used: Google Vertex (inherits from parent).
  • Tools used: Google Search Tool
Tools Used:
  • MCP Client Tool: (for internal knowledge ingestion and retrieval)
  • Google Search Tool: (for external fallback query resolution)
Note: For details on modifying the Tools, refer Tools Library section.
Agent Workflow Behavior Summary
  • The Knowledge Retrieval Agent receives a user query and searches the internal document set using a semantic RAG approach.
  • If a valid internal answer is found, it’s passed to the Search Agent for summarization.
  • If no relevant internal data is available or current, the Search Agent performs a web search.
  • The Search Agent then synthesizes a user-facing answer, clearly indicating the source — internal or web-based.
Sample Questions:
  • How do I reset my password in the system?
  • What is your refund policy for monthly subscriptions?