Vertex AI Gemini Agent Chatbot

A production-ready AI chatbot powered by Google Vertex AI Gemini with conversation memory and tool-calling capabilities.

Features

• 🧠 Persistent Memory: Conversation history stored in Firestore per session
• 🛠️ Tool Calling: Calculator, web fetch, and email stub tools
• 🚀 Production Ready: Deploys to GCP Cloud Run
• 💬 Real-time UI: Clean, responsive chat interface
• 🔒 Secure: Input validation, whitelisting, safe execution

Tools Available

1. Calculator: Safe mathematical expression evaluation
2. Web Fetch: HTTP requests to whitelisted public APIs
3. Email (Stub): Interface ready for Gmail API integration

Prerequisites

• Python 3.11+
• GCP Project with Vertex AI API enabled
• Firestore database created
• GCP credentials configured locally

Quick Start

1. Clone and Setup

cd vertex-gemini-agent-chatbot
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate
pip install -r requirements.txt

2. Configure Environment

cp .env.example .env
# Edit .env with your GCP project details

3. Authenticate with GCP

gcloud auth application-default login
gcloud config set project YOUR_PROJECT_ID

4. Run Locally

python app.py

Visit http://localhost:8080

Deployment to Cloud Run

1. Build and Deploy

gcloud run deploy vertex-gemini-chatbot \
  --source . \
  --platform managed \
  --region us-central1 \
  --allow-unauthenticated \
  --set-env-vars GCP_PROJECT_ID=your-project-id,GCP_LOCATION=us-central1,GEMINI_MODEL=gemini-2.0-flash-exp,FIRESTORE_COLLECTION=chat_sessions

2. Access Your App

The deployment will provide a URL like: https://vertex-gemini-chatbot-xxxxx-uc.a.run.app

Testing

cURL Examples

# Send a chat message
curl -X POST http://localhost:8080/api/chat \
  -H "Content-Type: application/json" \
  -d '{"session_id": "test-session-1", "user_message": "What is 25 * 47?"}'

# Reset conversation
curl -X POST http://localhost:8080/api/reset \
  -H "Content-Type: application/json" \
  -d '{"session_id": "test-session-1"}'

# Test web fetch tool
curl -X POST http://localhost:8080/api/chat \
  -H "Content-Type: application/json" \
  -d '{"session_id": "test-session-2", "user_message": "Fetch the current Bitcoin price from https://api.coindesk.com/v1/bpi/currentprice.json"}'

Architecture

User → Flask App → Vertex AI Gemini → Tool Execution → Firestore
         ↓                                ↓
    HTML/CSS/JS                    Calculator/WebFetch/Email

Security Notes

• Calculator: Only evaluates safe mathematical expressions
• Web Fetch: Whitelisted domains only
• Email: Stub implementation with OAuth placeholder
• Input validation on all endpoints
• Request size limits enforced

Email Tool Setup (Optional)

To enable the email tool:

1. Enable Gmail API in GCP Console
2. Create OAuth 2.0 credentials
3. Download credentials.json
4. Run the OAuth flow locally
5. Update shared/tools.py with credentials path
6. Uncomment email execution code

Environment Variables

• GCP_PROJECT_ID: Your GCP project ID
• GCP_LOCATION: Vertex AI location (e.g., us-central1)
• GEMINI_MODEL: Model name (default: gemini-2.0-flash-exp)
• FIRESTORE_COLLECTION: Collection name for chat history

License

MIT