SQL Assistant Crew is a Streamlit-powered, CrewAI-orchestrated multi-agent assistant that allows users to interact with a SQLite database using natural language. With a modular agent architecture for generation, review, and compliance checking, users can safely and intuitively run SQL queries—no manual SQL required. Checkout the blog for more details (https://towardsdatascience.com/a-multi-agent-sql-assistant-you-can-trust-with-human-in-loop-checkpoint-llm-cost-control/)

This app uses a CrewAI-based agent system to process each user query through a collaborative pipeline:

1. 🗣 Natural Language Input (User)
   Users type their question in everyday language using the Streamlit interface.

2. 🤖 SQL Generation Agent
   Converts the user prompt into an SQL query using RAG-enhanced schema context and user intent.

3. 🔍 Query Review Agent
   Refines and optimizes the SQL query for correctness and performance using the full database schema.

4. 🔐 Compliance Agent
   Validates the SQL for data safety, PII access, and policy violations.

5. ⚙️ Query Execution
   Runs the approved SQL query against the SQLite database and returns results to the user.

• 💬 Natural language SQL querying - Ask questions in plain English
• 🧠 4 RAG approaches to choose from - Keyword, FAISS Vector, ChromaDB Vector, or No RAG
• 🔬 Interactive RAG performance testing - Compare approaches with real-time metrics
• 🤖 Multi-agent system powered by CrewAI (Generation → Review → Compliance)
• 🗃️ Comprehensive ecommerce database with 20+ tables and sample data
• 🔄 One-click schema refresh support
• 🔐 Built-in query safety checks and compliance validation
• 💰 LLM cost transparency - track your token usage and costs

The system offers 4 different RAG approaches to optimize schema handling for different use cases:

1. No RAG (Full Schema)

   • Uses complete database schema for maximum context
   • Best for: Complex queries requiring full schema awareness
   • Trade-off: Higher token usage, slower processing

2. Keyword RAG

   • Custom keyword-based table selection
   • Best for: Fast processing with good accuracy
   • Benefits: 60-80% token reduction, <1ms response time

3. FAISS Vector RAG

   • Advanced semantic similarity search using FAISS
   • Best for: Complex semantic queries requiring deep understanding
   • Benefits: 75%+ token reduction, superior semantic matching

4. Chroma Vector RAG

   • Persistent vector database with ChromaDB
   • Best for: Production environments requiring consistent performance
   • Benefits: 75%+ token reduction, persistent storage, faster than FAISS

• Interactive comparison of all RAG approaches
• Real-time metrics showing table count, token reduction, and response time
• Query-specific recommendations based on performance results

• 📉 60-80% token reduction for schema context
• ⚡ 25% faster query generation with keyword RAG
• 🚀 Advanced semantic matching with vector RAG
• 💰 Lower costs - significant OpenAI API savings
• 🎯 Better accuracy with focused, relevant context

• Python 3.9+ (Python 3.11 recommended)
• OpenAI API key

# Clone the repository
git clone https://github.com/your-username/sql-assistant-crew.git
cd sql-assistant-crew

# Install dependencies
pip install -r requirements.txt

# Set up environment variables
echo "OPENAI_API_KEY=your_openai_api_key_here" > .env

# Initialize the ecommerce database
python -c "from utils.db_simulator import setup_ecommerce_db; setup_ecommerce_db(); print('✅ Database created')"

# Run the application
streamlit run app.py