This codelab walks you through building RAG-powered agents using Gemini File Search — Google’s fully-managed retrieval system built directly into the Gemini API. You’ll go from a baseline agent with no knowledge of your data, to a multi-agent system that can query your bespoke documents alongside live Google Search — all without writing a single line of embedding code.Documentation Index
Fetch the complete documentation index at: https://mintlify.com/derailed-dash/gemini-file-search-demo/llms.txt
Use this file to discover all available pages before exploring further.
Introduction
What this codelab covers and what you’ll build by the end
Prerequisites
Set up your Google Cloud project, clone the repo, and get your API key
What is RAG?
Understand why RAG matters and how Gemini File Search simplifies it
Gemini File Search
Explore how the File Search Store works and what document types it supports
What you’ll build
By the end of this codelab, you’ll have a multi-agent system that:- Uses Gemini File Search to answer questions about your bespoke documents
- Falls back to Google Search for general knowledge queries
- Is built on the Google Agent Development Kit (ADK) for clean, extensible structure
- Can be explored interactively through the ADK Web UI
Build the baseline agent
Start with a simple Gemini SDK agent that uses Google Search. Prove it cannot answer questions about your custom documents.
Create a File Search Store
Use a Jupyter notebook to create a Gemini File Search Store and upload your documents. No chunking, embedding, or vector DB needed.
Add RAG to your agent
Attach the File Search Store as a tool. Watch the agent correctly answer questions it previously had no knowledge of.
The approach
Basic SDK agent
A minimal agent using the raw Google Gen AI SDK with Google Search
File Search Store setup
Create and populate a File Search Store via Jupyter notebook
SDK RAG agent
Attach the File Search Store to your SDK agent for RAG capability
Multi-agent ADK system
Build a full multi-agent ADK system with both RAG and Google Search
This codelab is designed to run in Google Cloud Shell Editor or any local Python 3.12+ environment. Costs are minimal — completing the lab is expected to cost at most a few pennies.