General Prerequisites

• Python 3.x
• Google Cloud SDK
• Google Agent Development Kit (ADK)

Please refer to the individual agent directories for specific dependencies and configuration steps.

Available Agents

1. GCP Release Notes Agent

• Directory: gcp-releasenotes-agent-app/
• Description: An agent designed to answer questions about Google Cloud release notes. It connects to a MCP Toolbox for Databases service that queries a public BigQuery dataset.
• Features:
  • Demonstrates integration with a BigQuery-backed MCP Toolbox.
  • Includes instructions for deploying the toolbox service to Cloud Run.

For detailed setup and execution instructions, please see the GCP Release Notes Agent README.

2. Shop Search Agent

• Directory: shop-agent-app/
• Description: An agent that acts as a shopping assistant, using a tool to search for products in a catalog. It connects to a separate MCP server backed by Vertex AI Search for Retail.
• Features:
  • Illustrates how to connect an agent to a custom MCP server.
  • Provides a clear example of a retail or e-commerce use case.

For detailed setup and execution instructions, please see the Shop Search Agent README.

3. Restaurant Finder Agent

• Directory: restaurant-finder/
• Description: A conversational AI agent that helps users find restaurants based on a specific dish or ingredient, leveraging the Google Maps Platform API.
• Features:
  • Demonstrates how an agent can use external tools (Google Maps API) to answer user queries.
  • Provides real-time restaurant information.

For detailed setup and execution instructions, please see the Restaurant Finder Agent README.

4. Shopper's Concierge Agent

• Directory: shopper-concierge-demo/
• Description: An advanced shopping assistant that uses a sub-agent for research to provide more relevant product recommendations.
• Features:
  • Showcases a multi-agent architecture where a primary agent delegates tasks to a specialized sub-agent.
  • Includes a Gradio web interface for a complete user experience.

For detailed setup and execution instructions, please see the Shopper's Concierge Agent README.

5. Agentic RAG

This section includes agents that implement the Retrieval-Augmented Generation (RAG) pattern using different Google Cloud database services for vector search.

QnA Agent with AlloyDB

• Directory: RAG/rag-with-alloydb/
• Description: An agent that implements the RAG pattern using AlloyDB for PostgreSQL for vector search.
• Features:
  • Demonstrates using AlloyDB as a vector store for RAG.
  • Includes data ingestion scripts for populating the vector database.
  • Provides instructions for local execution and deployment to Vertex AI Agent Engine.

For detailed setup and execution instructions, please see the RAG with AlloyDB Agent README.

QnA Agent with BigQuery

• Directory: RAG/rag-with-bigquery/
• Description: An agent that implements the RAG pattern using BigQuery for vector search.
• Features:
  • Demonstrates using BigQuery as a vector store for RAG.
  • Includes data ingestion scripts.
  • Provides instructions for local execution and deployment to Vertex AI Agent Engine.

For detailed setup and execution instructions, please see the RAG with BigQuery Agent README.

QnA Agent with BigQuery Hybrid Search

• Directory: RAG/rag-with-bigquery-hybridsearch/
• Description: An agent that implements the RAG pattern using BigQuery Hybrid Search — a single retrieval step that combines BigQuery VECTOR_SEARCH() (semantic similarity) with SEARCH() (full-text keyword matching).
• Features:
  • Hybrid Search: Combines semantic and full-text keyword search using Reciprocal Rank Fusion (RRF).
  • Two Modes: Supports rrf (default balance) and pre_filter (keyword-strict) modes.
  • langchain-bigquery-hybridsearch: Uses an extension of BigQueryVectorStore for hybrid retrieval.
  • Data Ingestion: Includes scripts to populate the BigQuery vector store.

For detailed setup and execution instructions, please see the RAG with BigQuery Hybrid Search Agent README.

DEO Negation-Aware RAG with BigQuery

• Directory: RAG/deo-rag-with-bigquery/
• Description: An agent that implements DEO (Direct Embedding Optimization) for negation-aware retrieval using BigQuery as the vector store.
• Features:
  • Negation-Aware Retrieval: Handles queries with exclusion intent using gradient-based embedding optimization.
  • BigQuery Vector Search: Uses BigQuery for efficient vector search on optimized embeddings.
  • Query Decomposition: Automatically decomposes complex queries into positive and negative intents.

For detailed setup and execution instructions, please see the DEO RAG with BigQuery README.

QnA Agent with Spanner

• Directory: RAG/rag-with-spanner/
• Description: An agent that implements the RAG pattern using Google Cloud Spanner for vector search.
• Features:
  • Demonstrates using Spanner as a vector store for RAG.
  • Includes data ingestion scripts.
  • Provides instructions for local execution and deployment to Vertex AI Agent Engine.

For detailed setup and execution instructions, please see the RAG with Spanner Agent README.

QnA Agent with Vector Search 2.0

• Directory: RAG/rag-with-vectorsearch-2.0/
• Description: An agent that implements the RAG pattern using Vertex AI Vector Search 2.0 as a unified vector and document store, eliminating the need for a separate database like Firestore or AlloyDB.
• Features:
  • Unified Architecture: Simplifies RAG by storing embeddings and source data together.
  • Auto-Embeddings: Automatically generates embeddings using Vertex AI models during ingestion.
  • Hybrid Search: Combines semantic and keyword search with Reciprocal Rank Fusion (RRF).
  • Integrated Testing: Includes a Jupyter notebook for verifying the deployed agent on Agent Engine.

For detailed setup and execution instructions, please see the RAG with Vector Search 2.0 README.

QnA Agent with Gemini File Search

• Directory: RAG/rag-with-file-search/
• Description: An agent that implements the RAG pattern using Gemini File Search, a fully managed RAG system within the Gemini API ecosystem.
• Features:
  • Fully Managed RAG: Hands-off file search store management (chunking, embedding, storage).
  • Auto-Ingestion: Automatically indexes files uploaded through the ADK Web UI.
  • Session Isolation: Supports data isolation between users via session-based metadata filters.
  • MIME Type Patching: Robust handling of various file types with automated MIME type correction.

For detailed setup and execution instructions, please see the RAG with Gemini File Search README.

More RAG Agents

This repository contains additional RAG agent implementations that demonstrate integration with various Google Cloud services. Below is a list of other available RAG agents:

• RAG Engine with Managed DB:

  • Directory: RAG/rag-engine-with-managed-db/
  • Description: An agent that leverages the managed Vertex AI RAG Engine with its own fully managed database, eliminating the need to manage a separate Vector Search index.
  • README: RAG Engine with Managed DB README

• RAG Engine with Vector Search:

  • Directory: RAG/rag-engine-with-vectorsearch/
  • Description: An agent that uses the managed Vertex AI RAG Engine with a Vertex AI Vector Search index as its backend for efficient, scalable document retrieval.
  • README: RAG Engine with Vector Search README

• RAG with Vector Search and Datastore:

  • Directory: RAG/rag-with-vectorsearch-ds/
  • Description: An agent that uses Vertex AI Vector Search as the vector store and Firestore in Datastore mode as the document store.
  • README: RAG with Vector Search and Datastore README

• RAG with Vector Search and GCS:

  • Directory: RAG/rag-with-vectorsearch-gcs/
  • Description: An agent that uses Vertex AI Vector Search as the vector store and Google Cloud Storage (GCS) as the document store.
  • README: RAG with Vector Search and GCS README

6. Graph RAG

This section includes agents that implement the Graph Retrieval-Augmented Generation (Graph RAG) pattern.

Graph RAG with Spanner