Skip to content

pathrag-with-bigquery

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

pathrag-with-bigquery

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
assets
assets
 
 
data_ingestion
data_ingestion
 
 
pathrag_with_bigquery
pathrag_with_bigquery
 
 
README.md
README.md
 
 
requirements.txt
requirements.txt
 
 

README.md

PathRAG Agent with BigQuery Graph

This project demonstrates how to implement a PathRAG (Path-based Retrieval Augmented Generation) agent using the Agent Development Kit (ADK) with Google Cloud BigQuery as the storage backend.

It leverages the PathRAG library with the pathrag-bigquery storage plugin and LiteLLM for Gemini model integration.

Architecture

PathRAG Framework Architecture

Image Source: "PathRAG: Pruning Graph-based Retrieval Augmented Generation with Relational Paths" 
User Query
    |
    v
ADK Agent (Gemini 2.5 Flash)
    |  tool call
    v
pathrag_tool(query)
    |
    v
PathRAG.aquery(only_need_context=True)
    |-- Keyword Extraction (LLM)
    |-- Graph Search (BigQuery Property Graph)
    |-- Vector Search (BigQuery Vector Search)
    +-- Context assembly and return
    |
    v
ADK Agent generates final answer based on context

QueryParam(only_need_context=True) skips answer generation inside PathRAG, letting the ADK Agent's LLM generate the final answer from the retrieved context.

How It Works

  1. User sends a query to the ADK Agent.
  2. Agent calls pathrag_tool with the query.
  3. PathRAG processes the query:
    • Extracts keywords (high-level & low-level) using LLM.
    • Searches the BigQuery Property Graph (entities, relationships, paths).
    • Searches the BigQuery Vector Store (semantic similarity).
    • Combines results into structured context.
  4. Context is returned to the Agent (no LLM answer generation inside PathRAG).
  5. Agent generates the final answer using the retrieved context.

Project Structure

pathrag-with-bigquery/
├── pathrag_with_bigquery/            # ADK Agent directory
│   ├── __init__.py
│   ├── agent.py                     # ADK Agent definition (root_agent)
│   ├── prompt.py                    # Agent system instructions
│   ├── tools.py                     # pathrag_tool - context retrieval via PathRAG
│   └── .env.example                 # Environment variables template
├── data_ingestion/                  # Data ingestion directory
│   └── insert.py                    # Script to ingest documents
├── requirements.txt                 # Project dependencies
└── README.md

Key Files

File Description
pathrag_with_bigquery/agent.py root_agent definition using Gemini 2.5 Flash and pathrag_tool
pathrag_with_bigquery/tools.py pathrag_tool function, extracts context from PathRAG
pathrag_with_bigquery/prompt.py System instruction guiding the Agent to answer based on tool-retrieved context
data_ingestion/insert.py Script to ingest documents into the PathRAG Knowledge Graph

Storage Backend

This project uses Google Cloud BigQuery for scalable, serverless storage. Tables and Property Graph are automatically created by pathrag-bigquery on first use via lazy initialization (_ensure_schema()).

Component Backend
KV Storage BigQueryKVStorage
Vector Storage BigQueryVectorDBStorage
Graph Storage BigQueryGraphStorage

Prerequisites

Before you begin, ensure you have the following tools installed:

1. Configure your Google Cloud project

First, authenticate with Google Cloud:

gcloud auth application-default login

Next, set up your project and enable the necessary APIs:

export PROJECT_ID=$(gcloud config get-value project)

gcloud services enable \
  bigquery.googleapis.com \
  aiplatform.googleapis.com

2. Create a BigQuery Dataset

Create a BigQuery dataset using the gcloud CLI.

# Set environment variables
export BIGQUERY_PROJECT=$PROJECT_ID
export BIGQUERY_DATASET="pathrag"
export BIGQUERY_LOCATION="us-central1"

# Create the BigQuery dataset
bq --location=$BIGQUERY_LOCATION mk \
  --dataset \
  --description="PathRAG Dataset" \
  ${BIGQUERY_PROJECT}:${BIGQUERY_DATASET}

3. Set Environment Variables

Copy the example file and edit it:

cp pathrag_with_bigquery/.env.example pathrag_with_bigquery/.env
export GOOGLE_CLOUD_PROJECT="your-project-id"
export GOOGLE_CLOUD_LOCATION="us-central1"
export GOOGLE_GENAI_USE_VERTEXAI="1"
export BIGQUERY_PROJECT="your-project-id"
export BIGQUERY_DATASET="pathrag"

Setup

1. Install Dependencies

This project uses uv to manage the Python virtual environment and package dependencies.

Create and activate the virtual environment:

# Create the virtual environment
uv venv

# Activate the virtual environment
source .venv/bin/activate

Install dependencies:

uv pip install -r requirements.txt

2. Data Ingestion

First, load the environment variables from the .env file:

source pathrag_with_bigquery/.env

Ingest documents into the PathRAG Knowledge Graph.

# Ingest sample documents (Apple, Steve Jobs, Google)
python data_ingestion/insert.py --sample

# Or ingest your own document
python data_ingestion/insert.py --file your_document.txt

3. Run the Agent

You can run the agent using either the command-line interface or a web-based interface.

Using the Command-Line Interface (CLI)

adk run pathrag_with_bigquery

Using the Web Interface

adk web

Screenshot:

pathrag-with-bigquery
Figure 1. PathRAG with BigQuery - ADK Web UI
pathrag-with-bigquery-storages
Figure 2. PathRAG with BigQuery - Storages 		pathrag-with-bigquery-adk_log
Figure 3. PathRAG with BigQuery - ADK Log

References