🤖 LeoBot — AI Chat Assistant for Businesses & Users

LeoBot is a FastAPI-based AI chatbot platform for intelligent, real-time conversations across multiple channels — including websites, Facebook Messenger, and Zalo Official Accounts. It integrates seamlessly with LEO CDP (Customer Data Platform) to serve both admins and end-users, delivering contextual, personalized answers powered by RAG (Retrieval-Augmented Generation) and advanced AI models like Google Gemini.

---

🌐 Live Chatbot Demo

Try the chatbot in action: 👉 https://leobot.leocdp.com

LeoBot uses the Google Gemini API for natural, context-aware responses.

---

🚀 Features

• Multi-channel support: Works with Facebook Messenger and Zalo OA.
• Gemini-powered intelligence: Uses Google Gemini API for high-quality understanding and generation.
• RAG-based reasoning: Combines knowledge retrieval with semantic memory.
• FastAPI backend: Lightweight, async, and production-ready.
• Redis rate limiting: Prevents spam and message floods.
• Custom personas: Supports user profiles, roles, and chat touchpoints.
• Prebuilt frontend demos: Jinja2 templates for quick UI testing and embedding.

---

🧠 Architecture Overview

(Coming soon — overview diagram and explanation of key modules.)

---

🧩 Key Components

RAGAgent

Handles message understanding, context retrieval, and Gemini-based response generation.

Webhooks

• /fb-webhook — Facebook Messenger
• /zalo-webhook — Zalo Official Account

Redis Rate Limiting

Uses a sorted-set time window to prevent excessive messaging per user.

---

🛠️ Setup and Installation

1. Clone the Repository

git clone https://github.com/trieu/leo-bot
cd leo-bot

2. Start PostgreSQL + pgvector (via Docker)

Make sure Docker CLI is installed:

docker --version

Then start the database:

./dockers/pgsql/start_pgsql_pgvector.sh

This script:

• Launches a PostgreSQL 16 container (pgsql16_vector)
• Mounts a persistent volume (pgdata_vector)
• Enables pgvector and postgis extensions
• Creates the customer360 database and schema
• Handles collation version fixes automatically

To reset the database, run:

./dockers/pgsql/start_pgsql_pgvector.sh --reset-db

You can connect manually:

psql -h localhost -U postgres -d customer360

---

3. Install Dependencies

pip install -r requirements.txt

4. Configure Environment

Create a .env file or edit main_config.py:

# Core LEO BOT
LEOBOT_DEV_MODE=true
HOSTNAME=leobot.example.com
POSTGRES_URL=postgresql://postgres:password@localhost:5432/customer360

# Google API
GOOGLE_APPLICATION_CREDENTIALS= 
GEMINI_API_KEY= 

# Set Redis HOST and PORT 
REDIS_USER_SESSION_HOST=localhost
REDIS_USER_SESSION_PORT=6379

# Set SMTP email information
SMTP_HOST=""
SMTP_PORT=587    
SMTP_USER="" 
SMTP_PASSWORD=""  
USE_TLS=true 

# in dev, disable HTTPS check certification
PYTHONHTTPSVERIFY=1

# Keycloak for leobot client at DEV mode
KEYCLOAK_ENABLED=false
KEYCLOAK_URL=https://leoid.example.com
KEYCLOAK_REALM=master
KEYCLOAK_CLIENT_ID=leobot
KEYCLOAK_CLIENT_SECRET=""
KEYCLOAK_CALLBACK_URL=https://leobot.example.com/_leoai/sso/callback
KEYCLOAK_VERIFY_SSL=false

# Set your Facebook Page Access here
FB_VERIFY_TOKEN=""
FB_PAGE_ACCESS_TOKEN=""

Gemini API setup:

• Get your API key at Google AI Studio
• For translation and related APIs, set up credentials in the Google Cloud Console

---

5. Python Environment (Ubuntu Example)

sudo apt install python-is-python3 python3.10-venv
python -m venv env
source env/bin/activate
pip install -r requirements.txt

After installation, refresh your shell.

---

6. Run LeoBot

Production mode:

./start_app.sh

Development mode:

./start_dev.sh

LeoBot will run at 0.0.0.0:8888. Open your browser and visit your configured HOSTNAME to test.

---

🌐 API Endpoints

Endpoint	Method	Description
/_leoai/ask	POST	Main chatbot endpoint
/_leoai/is-ready	GET/POST	Gemini API readiness check
/_leoai/fb-webhook	GET/POST	Facebook Messenger webhook
/_leoai/zalo-webhook	POST	Zalo OA webhook
/_leoai/ping	GET	Basic health check
/_leoai/visitor-info	GET	Retrieve visitor info from Redis

---

🧰 Developer Notes

• Built on FastAPI with full async I/O.
• Message context stored in Redis.
• Embeddings via SentenceTransformer or Gemini Embeddings.
• Compatible with pgvector and other vector databases.

To extend LeoBot:

• Add webhook routes for new channels (Telegram, LINE, etc.)
• Create custom response modules in rag_agent
• Integrate new LLM APIs or plugins

---

🧪 Testing

Run automated tests:

pytest

Or test manually:

curl -X POST http://localhost:8000/ask \
  -H "Content-Type: application/json" \
  -d '{"visitor_id": "demo", "question": "Hello!", "persona_id": "test"}'

---

🧭 Author & Resources

Author: Trieu Nguyen YouTube: @bigdatavn Demo: https://leobot.leocdp.com

---

🌱 Future Roadmap

• Support Telegram, LINE, and WhatsApp
• Add knowledge-graph search (PostgreSQL + pgvector)
• Streaming chat via SSE/WebSocket
• AI analytics dashboard for admins
• Plugin SDK for external integrations

---

📜 License

MIT License — free to use, modify, and share. Attribution is appreciated but not required..

---

💡 Vision

LeoBot embodies the union of Dataism and AI pragmatism — an assistant that connects people and data through natural conversation. It’s not just automation; it’s augmentation — amplifying human understanding through intelligent dialogue.