Sav6 - AI-Powered Chat Agent with Workflow Builder

A sophisticated data engineering and analytics platform featuring an AI chat agent that can create workflows, analyze data, and automate business processes.

🚀 Features

Chat Agent

• Mission-Critical Accuracy Mode - Never guesses or assumes, always verifies
• Four Operational Modes:
  • 🌟 Planning - Strategic thinking and approach planning
  • 🔨 Building - Create workflows, datasets, and artifacts
  • 📊 Analyzing - Data analysis and insights
  • ▶️ Running - Execute existing workflows
• Streaming Responses with Server-Sent Events (SSE)
• File Upload Support for data analysis
• Conversation Persistence with SQLite
• Policy Enforcement - Mandatory, not optional

Workflow Builder

• Visual Graph Editor with React Flow
• 17 Node Types from SAV system:
  • Filter, Dedupe, Summarize, Join
  • Pivot, Unpivot, Transform, Time Series
  • Upload, Destination, Bot, Add
  • External connectors (Airtable, GitHub, Facebook, Atlassian)
• Agent-Generated Workflows - Chat creates workflows automatically
• Save/Load to SQLite database

Tech Stack

• Frontend: React 19, Tailwind CSS 4, React Flow, Lucide Icons
• Backend: Express.js, TypeScript, sql.js (pure JavaScript SQLite)
• AI: Google Gemini API with streaming
• Database: SQLite with full schema for workflows, conversations, logs

📦 Installation

cd sav6
npm install

⚙️ Configuration

Option 1: Environment Variable

Create .env.local:

GEMINI_API_KEY=your_api_key_here

Option 2: Settings UI

1. Start the application
2. Navigate to Settings
3. Enter your Gemini API key
4. Click Save

Get your API key from Google AI Studio

🎯 Usage

Start Development Servers

Option A: Run both together (Recommended)

npm run dev:all

Option B: Run separately

Terminal 1 - Frontend:

npm run dev

Terminal 2 - Backend:

npm run server

Access the Application

• Frontend: http://localhost:3000
• Backend API: http://localhost:3001
• Database: Auto-created at data/database.sqlite

💬 Using the Chat Agent

1. Navigate to Chat (first item in left navigation)
2. Select a mode: Planning, Building, Analyzing, or Running
3. Ask questions or describe what you need:
   • "Create a workflow to filter customer data from California"
   • "Analyze sales trends for Q1 2024"
   • "Set up a data pipeline for customer segmentation"
4. Upload files for data analysis
5. Agent will:
   • Ask clarifying questions
   • Create workflows with proper node configurations
   • Generate datasets and markdown documentation
   • Provide mission-critical, accurate responses

🔧 API Endpoints

Chat

• POST /api/chat/message - Send message with streaming response
• GET /api/chat/history/:conversationId - Get conversation history
• GET /api/chat/conversations - List all conversations

Workflows

• GET /api/workflow - List all workflows
• GET /api/workflow/:id - Get workflow details
• POST /api/workflow - Create workflow
• PUT /api/workflow/:id - Update workflow
• DELETE /api/workflow/:id - Delete workflow

Upload

• POST /api/upload - Upload single file
• POST /api/upload/multiple - Upload multiple files

Settings

• GET /api/settings - Get all settings
• GET /api/settings/:key - Get specific setting
• POST /api/settings - Set setting
• DELETE /api/settings/:key - Delete setting

📁 Project Structure

sav6/
├── server/               # Express backend
│   ├── index.ts         # Server entry point
│   ├── database.ts      # SQLite database
│   ├── routes/          # API routes
│   │   ├── chat.ts      # Chat endpoints
│   │   ├── workflow.ts  # Workflow CRUD
│   │   ├── upload.ts    # File uploads
│   │   └── settings.ts  # Settings management
│   └── services/        # Business logic
│       ├── llm.ts       # Gemini integration
│       └── prompts.ts   # System prompts
├── src/                 # React frontend
│   └── views/           # Page components
│       ├── Chat.tsx     # Chat interface
│       └── Settings.tsx # Settings page
├── data/                # Runtime data
│   ├── database.sqlite  # SQLite database
│   └── uploads/         # Uploaded files
└── md-library/          # Policy/skills/knowledge

🗄️ Database Schema

Tables

• workflows - Workflow metadata
• nodes - Workflow nodes with configs
• edges - Node connections
• conversations - Chat conversation metadata
• chat_messages - Individual messages
• execution_logs - Workflow execution history
• settings - API keys and configuration

🎨 Design Philosophy

• Swiss Minimalism - Clean, grid-based layouts
• Brand Color: #6453FF (Savant Purple)
• Mission-Critical UX - No ambiguity, always clear
• Progressive Disclosure - Show complexity when needed
• Workflow-First - Everything revolves around workflows

🛠️ Development

Type Checking

npm run lint

Build for Production

npm run build

Clean Build Artifacts

npm run clean

📝 Use Cases

1. One-off Requests - Quick data analysis via chat
2. Exploration - Discover patterns and insights
3. First Deliverable - Create initial workflows
4. Repeat Workflows - Save and reuse workflows
5. Full Automation - Schedule and monitor workflows
6. Documentation - Generate MD files for processes

🐛 Troubleshooting

Port Already in Use

Change ports in:

• Frontend: vite.config.ts (default: 3000)
• Backend: server/index.ts (default: 3001)

Database Errors

Ensure data directory exists:

mkdir -p data/uploads

API Key Not Working

1. Verify key in Settings page
2. Check browser console for errors
3. Ensure backend server is running
4. Test with /api/health endpoint

📖 Learn More

• React Flow Documentation
• Gemini API Documentation
• SQLite Documentation
• Express.js Documentation

🤝 Contributing

This is a proof-of-concept for the Savant Labs platform. The system demonstrates:

• AI-powered workflow generation
• Mission-critical agent behavior
• Real-time streaming chat
• Visual workflow editing
• Local-first data storage

📄 License

Proprietary - Savant Labs

---

Built with ❤️ for Data Engineers and Analytics Teams