oss-data-analyst - Open Source AI Data Science Agent [Reference Architecture]

oss-data-analyst is an intelligent AI agent that converts natural language questions into SQL queries and provides data analysis. Built with the Vercel AI SDK, it features multi-phase reasoning (planning, building, execution, reporting) and streams results in real-time.

Note: This is a reference architecture. The semantic catalog and schemas included are simplified examples for demonstration purposes. Production implementations should use your own data models and schemas.

Features

• Multi-Phase AI Agent: Planning → Building → Execution → Reporting workflow
• Real-time Streaming: Live updates during query processing
• Smart Data Analysis: Automated insights and visualizations
• SQL Validation: Syntax checking and security policy enforcement
• Natural Language: Ask questions in plain English
• Modern UI: Built with Next.js, React, and TailwindCSS
• Extensible Tools: Easy to add custom tools and capabilitiets

Quick Start

Prerequisites

• Node.js 20.19.3+
• pnpm 8.15.0+
• AI Gateway API key

Installation

1. Clone the repository

   git clone https://github.com/vercel/oss-data-analyst.git
   cd oss-data-analyst

2. Install dependencies

   pnpm install

3. Set up environment variables

   cp env.local.example .env.local

   Edit .env.local and add your Vercel AI Gateway key

4. Initialize the database

   pnpm initDatabase

   This creates a SQLite database with sample data (Companies, People, Accounts)

5. Run the development server

   pnpm dev

6. Open your browser Navigate to http://localhost:3000

Build for Production

pnpm build
pnpm start

Sample Schema

This repository includes a sample database schema with three main entities to demonstrate oss-data-analyst's capabilities:

Companies

Represents organizations in your database. Each company has:

• Basic information (name, industry, employee count)
• Business metrics (founded date, status)
• Example: Technology companies, Healthcare organizations, etc.

Accounts

Represents customer accounts or subscriptions tied to companies. Each account includes:

• Account identification (account number, status)
• Financial metrics (monthly recurring value, contract details)
• Relationship to parent company
• Example: Active subscriptions with monthly values ranging from $10k-$50k

People

Represents individual employees or contacts within companies. Each person has:

• Personal information (name, email)
• Employment details (department, title, salary)
• Relationship to their company
• Example: Engineers, Sales representatives, Managers across different departments

How It Works

oss-data-analyst uses a multi-phase agentic workflow:

1. Planning Phase

   • Analyzes natural language query
   • Searches semantic catalog for relevant entities
   • Identifies required data and relationships
   • Generates execution plan

2. Building Phase

   • Constructs SQL query from plan
   • Validates syntax and security policies
   • Optimizes query structure
   • Finds join paths between tables

3. Execution Phase

   • Estimates query cost
   • Executes SQL against database
   • Handles errors with automatic repair
   • Streams results

4. Reporting Phase

   • Formats query results
   • Generates visualizations (charts, tables)
   • Provides natural language explanations
   • Performs sanity checks on data

Extending oss-data-analyst

Customizing Prompts

Modify system prompts in src/lib/prompts/:

• planning.ts - Planning phase behavior
• building.ts - SQL generation logic
• execution.ts - Query execution handling
• reporting.ts - Results interpretation

Example Queries

Try asking oss-data-analyst (using the sample database):

• "How many companies are in the Technology industry?"
• "What is the average salary by department?"
• "Show me the top 5 accounts by monthly value"
• "Which companies have the most employees?"
• "What is the total revenue for Active accounts?"
• "How many people work in Engineering?"

Using with Production Databases

The default setup uses SQLite for demonstration. To use with Snowflake or other databases:

1. Update src/lib/oss-data-analyst-agent-advanced.ts to import from ./tools/execute instead of ./tools/execute-sqlite
2. Configure your database credentials in .env.local
3. Update the semantic catalog in src/lib/semantic/ with your schema definitions

Troubleshooting

Database Not Found

• Run pnpm initDatabase to create and seed the database
• Check that data/oss-data-analyst.db exists

AI Gateway API Errors

• Verify your API key is valid in .env.local
• Check API rate limits and credits

Build Errors

• Run pnpm install to update dependencies
• Check TypeScript errors with pnpm run type-check
• Clear .next folder and rebuild