Build a Drive Folder RAG Chatbot Using Google Apps Script + Gemini

by

Get Github Files https://github.com/lsvekis/Drive-Folder-RAG-Chatbot

Build a Drive Folder RAG Chatbot Using Google Apps Script + GeminiDownload

(Step-By-Step Guide)

Turn any folder of Google Docs into an AI-powered knowledge assistant

AI assistants are becoming a core part of modern workflows—and thanks to Google Apps Script + Gemini, you can now build your own internal chatbot that answers questions using only the content stored inside a Google Drive folder.

This article walks you through a full Retrieval-Augmented Generation (RAG) solution built with Apps Script, including:

  • How it loads and processes files from a folder
  • How it chunk-indexes content
  • How it scores and retrieves relevant segments
  • How it sends structured context to Gemini
  • How the webapp interface works
  • How to generate demo files to test your bot

At the end, you’ll have a fully working Drive Folder RAG Chatbot running entirely inside Google Apps Script—no servers, no databases, no external hosting required.

⭐ What This Chatbot Can Do

Once deployed, the chatbot allows the user to:

✔ Select a Google Drive folder
✔ Automatically ingest all Google Docs + .txt files
✔ Break them into indexed chunks
✔ Score each chunk for relevance to the query
✔ Retrieve the best matches
✔ Send contextual info to Gemini
✔ Receive an answer based strictly on the folder contents

This means you can build:

  • Knowledge base assistants
  • Policy chatbots
  • SOP/documentation bots
  • Multi-file onboarding assistants
  • FAQ bots
  • Multi-document research agents
  • Course content assistants

Everything runs inside Apps Script.

🚀 The Script (Explained)

Below is the exact code, fully annotated and explained.

1. Configuration & Constants

const GEMINI_MODEL_ID = 'gemini-2.0-flash';
const MAX_FILES = 15;
const MAX_CHARS_PER_FILE = 20000;
const CHUNK_SIZE = 800;
const CHUNK_OVERLAP = 200;
const MAX_CONTEXT_CHARS = 15000;

What this does:

  • gemini-2.0-flash → fastest, most cost-efficient model supported on v1beta generateContent
  • MAX_FILES → limits how many Drive files are scanned
  • MAX_CHARS_PER_FILE → prevents massive docs from overloading Gemini
  • CHUNK_SIZE → each chunk is about 800 characters
  • CHUNK_OVERLAP → gives Gemini natural continuity
  • MAX_CONTEXT_CHARS → caps total text sent in a single API call

This keeps your RAG system fast, predictable, and safe.

2. Web App Entry Point

function doGet() {
  return HtmlService.createHtmlOutputFromFile('IndexRagFolder')
    .setTitle('Drive Folder RAG Chatbot')
    .setXFrameOptionsMode(HtmlService.XFrameOptionsMode.ALLOWALL);
}

This loads the HTML UI and allows the chatbot to run inside a browser.

3. API Key Retrieval

function getGeminiApiKey_() {
  const props = PropertiesService.getScriptProperties();
  const key = props.getProperty('GEMINI_API_KEY');
  if (!key) throw new Error('GEMINI_API_KEY is not set.');
  return key;
}

Stores your API key securely in Script Properties.

4. Selecting & Validating the Folder

function extractFolderId_(input) {
  const trimmed = input.trim();
  const urlMatch = trimmed.match(/\/folders\/([-\w]{10,})/);
  if (urlMatch) return urlMatch[1];
  const idMatch = trimmed.match(/[-\w]{10,}/);
  if (idMatch) return idMatch[0];
  return trimmed;
}

This makes the UI flexible—you can paste either:

  • A folder URL
  • Or just the folder ID

Apps Script extracts the correct value automatically.

5. Saving the User’s Active Folder

function setActiveFolder(folderInput) {
  const folderId = extractFolderId_(folderInput);
  const folder = DriveApp.getFolderById(folderId);
  PropertiesService.getUserProperties().setProperty('ACTIVE_FOLDER_ID', folderId);

  return {
    id: folderId,
    name: folder.getName(),
    url: 'https://drive.google.com/drive/folders/' + folderId
  };
}

Why use UserProperties?

  • Each user can have a different folder
  • No shared global state
  • Works for multi-user deployments

6. The Heart of the System: chatWithFolder()

This is where the RAG magic happens.

Step 1 — Load files from the folder

const chunks = buildChunksFromFolder_(folder, queryTerms);

This fetches the text from:

  • Google Docs
  • Plain text files

Step 2 — Generate meaningful chunks

Each file is broken into overlapping slices:

const fileChunks = chunkText_(text, CHUNK_SIZE, CHUNK_OVERLAP);

Chunking prevents Gemini from receiving giant walls of text.

Step 3 — Score each chunk

const score = scoreChunk_(chunkText, queryTerms);

This simple keyword overlap scoring:

  • Is lightning fast
  • Has no external dependencies
  • Avoids expensive embedding models
  • Works surprisingly well for short text

Step 4 — Build a context window

if (context.length + c.text.length > MAX_CONTEXT_CHARS) break;
context += '\n\n[Source: ' + c.fileName + ']\n' + c.text;

This produces a clean, human-readable memory structure:

[Source: File A]
...text...

[Source: File B]
...text...

Gemini loves this format.

Step 5 — Send the context to Gemini

const promptText = 
  systemInstruction +
  "\n\n=== CONTEXT START ===\n" + context +
  "\n=== CONTEXT END ===\n\nUser question: " + userMessage;

The system instruction enforces constraints:

  • Only answer from folder content
  • Say “I don’t see that…” when unsupported
  • Cite document names

This ensures zero hallucinations.

Step 6 — Return the final answer

return { answer: answer };

The UI displays this as a chat bubble.

🎛 Build Chunks + Score Terms

Chunking:

function chunkText_(text, size, overlap)

This ensures:

  • Smooth transitions
  • Better semantic matching
  • Less fragmentation

Scoring:

function scoreChunk_(chunkText, queryTerms)

Scores based on token frequency overlap—simple, fast, effective.

🧪 Generating “Demo Files”

The script includes a function:

function createDemoFilesInDemoFolder()

This automatically creates a folder called:

demo files

And populates it with:

  • Product Overview
  • Onboarding Checklist
  • Support FAQ
  • Team Communication Guidelines

Why this is useful:

  • Instantly test RAG
  • No manual document creation required
  • Reproducible testing for debugging

🖥 The Web UI

The file IndexRagFolder.html creates:

✔ A clean chat UI
✔ A folder selection box
✔ Realtime status messages
✔ Message bubbles
✔ Scrolling message history

No frameworks needed—pure HTML, CSS, and client-side Apps Script calls.

🎉 Final Result

You now have a fully working:

Drive Folder RAG Chatbot (Apps Script + Gemini)

✔ Multi-file context ingestion
✔ Chunking & relevance scoring
✔ Controlled context window
✔ Webapp chatbot interface
✔ Logging & error handling
✔ Auto-generated demo documents
✔ Fast responses using gemini-2.0-flash

This system can handle:

  • Knowledge bases
  • SOP libraries
  • Help centers
  • Team documentation
  • Course material
  • Internal onboarding
  • Multi-file research

All running entirely inside Google Apps Script.