How to Create an AI Agent for SQL Queries with CrewAI and GibsonAI

This guide explains how to build an AI Agent using CrewAI for orchestrating SQL queries and GibsonAI for handling data storage and CRUD operations via its Data API.

What You’ll Build

• A CrewAI Agent that uses the GibsonAI Data API to read and write data.
• You will define tables in GibsonAI, and CrewAI will use its API to query or insert records.
• The example provided demonstrates storing sales contact information in GibsonAI.

Key Concept

• GibsonAI exposes a REST Data API for all created tables.
• CrewAI can query and perform CRUD operations directly via this API, making it a powerful backend for AI agents.
• The ability to execute SQL queries via GibsonAI’s /query endpoint.

GitHub Repo Link: Sales Contact Finder (CrewAI + GibsonAI)

Prerequisites

Before you begin, ensure you have:

1. A GibsonAI Account – Sign up at https://app.gibsonai.com.
2. A GibsonAI API Key – Create a project in GibsonAI and copy the API key from the Connect tab.
3. Python 3.9+ installed.
4. OpenAI API Key – Get one here.
5. Serper.dev API Key (if using web scraping/search features).

1. Generate Your Database Schema in GibsonAI

   Use the following prompt in GibsonAI to create the schema:

   I want to create a sales contact aggregator agent.
   Generate a “sales_contact” table with fields (company_id, name, title, linkedin_url, phone, email).
   Also create a “sales_company” table with fields (name).
   All string fields, except name, are nullable.

   Click Deploy and copy the API Key.

   ---

2. Clone the Sales Contact Finder Example

   This example lives in the awesome-gibson repo. Clone it:

   git clone https://github.com/GibsonAI/awesome-gibson.git
   cd awesome-gibson/sales_contact_finder

3. Configure Your Environment

   Copy and edit the .env file:

   cp .env.example .env

   Fill in:

   GIBSONAI_API_KEY=your_project_api_key
   SERPER_API_KEY=your_serper_api_key
   OPENAI_API_KEY=your_openai_api_key

   ---

4. Create and Activate Virtual Environment

   source .venv/bin/activate  # For Windows: .venv\Scripts\activate

   ---

5. Install Dependencies

   uv pip sync pyproject.toml

   ---

6. Implement CrewAI Tool for SQL Operations

   CrewAI will communicate with GibsonAI’s Data API for CRUD operations. Below is an example ContactStorageTool:

   import json
   import os
   import requests
   from dotenv import load_dotenv
   from pydantic import Field
   from crewai.tools import BaseTool
   
   load_dotenv()  # Load environment variables from .env
   
   class ContactStorageTool(BaseTool):
       name: str = "ContactStorageTool"
       description: str = """
       Saves contact information in a GibsonAI database using the hosted API.
       Expected payload format:
       {"company_name": "Company Name", "contacts": [{"name": "Name", "title": "Title",
       "linkedin_url": "LinkedIn URL", "phone": "Phone", "email": "Email"}]}
       """
   
       api_base_url: str = Field(description="The base URL of the GibsonAI API")
       api_key: str = Field(description="The API key associated with your GibsonAI project")
   
       def __init__(self):
           self.api_base_url = "https://api.gibsonai.com/v1/-"
           self.api_key = os.getenv("GIBSONAI_API_KEY")
           if not self.api_key:
               raise ValueError("Missing GIBSONAI_API_KEY environment variable")
           super().__init__()
   
       def _run(self, contact_info: str) -> str:
           try:
               contact_data = json.loads(contact_info) if isinstance(contact_info, str) else contact_info
               company_name = contact_data["company_name"]
               contacts = contact_data["contacts"]
   
               # Insert company
               company_payload = {"name": company_name}
               response = requests.post(
                   f"{self.api_base_url}/sales-company",
                   json=company_payload,
                   headers={"X-Gibson-API-Key": self.api_key},
               )
               response.raise_for_status()
               company_id = response.json()["id"]
               print(f"Posted company: {response.status_code}")
   
               # Insert contacts
               for contact in contacts:
                   contact_payload = {
                       "company_id": company_id,
                       "name": contact["name"],
                       "title": contact["title"],
                       "linkedin_url": contact["linkedin_url"],
                       "phone": contact["phone"],
                       "email": contact["email"],
                   }
                   response = requests.post(
                       f"{self.api_base_url}/sales-contact",
                       json=contact_payload,
                       headers={"X-Gibson-API-Key": self.api_key},
                   )
                   print(f"Posted contact {contact['name']}: {response.status_code}")
           except Exception as e:
               return f"Failed to post contact: {str(e)}"

   ---

7. Run Your Crew

   Run:

   python main.py run

   The crew will:

   • Gather data (e.g., sales contacts).
   • Use the GibsonAI Data API to store the results.

8. Check Your Data

   Go to the GibsonAI Dashboard to see:

   • Sales Company and Sales Contact tables.
   • The data stored by the CrewAI agent.