SQLagent

SQLagent is a flexible agentic framework for interacting with SQL databases and answering natural language questions. It supports both general question-answering (using LangGraph and SmolAgents) and code generation tasks (such as plotting graphs from database queries).

Features

• agent.py: Contains both LangGraph and SmolAgents-based agents for general question answering and retrieval-augmented generation (RAG).
• transform_agent.py: Uses SmolAgents' CodeAgent for code generation tasks, such as plotting graphs for monthly bookings.
• conf.py: Central configuration for model endpoints, model names, and prompt templates.

Setup

1. Create a Virtual Environment

python -m venv .venv

Activate the virtual environment:

• On Windows:

  .venv\Scripts\activate

• On Unix/Mac:

  source .venv/bin/activate

2. Install Requirements

pip install -r requirements.txt

3. Environment Variables

Create a .env file in the project root with the following variables:

GROQ_API_KEY=your_groq_api_key
POSTGRES_URI=your_postgres_connection_uri
CHAT_API_KEY=your_ollama_or_openai_key
SQL_MODEL_API_KEY=your_sql_model_api_key

Other optional variables can be found in the provided .env sample.

4. Configuration

Edit conf.py to set the base URLs and model names for your agents and LLMs. For example:

BASE_URL = "http://localhost:11434"
CODE_MODEL = "llama3.3:70b"
SQL_BASE_URL = "https://openrouter.ai/api/v1"
SQL_MODEL = "meta-llama/llama-3.3-70b-instruct:free"

Adjust these values according to your deployment and model preferences.

Usage

Running the General Agent

• LangGraph agent:
  By default, running python agent.py will use the LangGraph agent via the run_lang() function.

  python agent.py

• SmolAgents agent:
  To use the SmolAgents agent instead, open agent.py and comment out the run_lang() line in the if __name__ == "__main__": block, then uncomment the run_smol() line. Example:

  # run_lang()
  run_smol()

  Then run:

  python agent.py

  This will prompt for user input and answer general questions about your database or markdown documentation.

Running the Code Agent

• SmolAgents CodeAgent (code generation, e.g., plotting):

  python transform_agent.py

  This agent is suited for tasks like generating code to plot graphs from SQL data.

• Streamlit UI for SmolAgents CodeAgent

  streamlit run app.py

Notes

• Ensure your .env variables are set correctly for database and API access.
• You can switch between different agent types and models by editing agent.py, transform_agent.py, and conf.py.
• For advanced configuration, refer to the comments in each file.