Quickstart

Time estimate: ~15 minutes. You will have a fully working DualMind Lab instance with chat, arena comparisons, and voting.

Prerequisites

.NET 8 SDK

Download from dotnet.microsoft.com

dotnet --version
# Should output: 8.0.x

Node.js 18+

Download from nodejs.org

node --version && npm --version

Supabase project

Create a free project at supabase.com. You need:

Project URL (e.g., https://xxx.supabase.co)
Anon key (Settings > API)
Service role key (Settings > API)
JWT secret (Settings > API > JWT Secret)

Save the JWT Secret immediately — you cannot retrieve it later without resetting it.

Groq API key

Get a free key at console.groq.com. This is the primary AI provider.

Groq’s free tier gives you 30 requests/min and up to 800 tokens/sec — plenty for development.

Setup

Clone repositories

git clone https://github.com/HarshBhanushali07/DualMind_Back.git
git clone https://github.com/HarshBhanushali07/DualMind_UI.git
git clone https://github.com/HarshBhanushali07/DualMind_Admin-UI.git

Configure environment variables

Backend (.env)
Frontend (config.js)

Create .env in DualMind_Back/src/DualMind.API/:

SUPABASE_URL=https://your-project.supabase.co
SUPABASE_SERVICE_ROLE_KEY=eyJ...your-service-role-key
SUPABASE_KEY=eyJ...your-anon-key
JWT_SECRET=your-jwt-secret-from-supabase
GROQ_API_KEY=gsk_...your-groq-api-key

Never commit .env files to Git. The .gitignore already excludes them.

Edit DualMind_UI/config.js:

window.DUALMIND_CONFIG.supabase.url = 'https://your-project.supabase.co';
window.DUALMIND_CONFIG.supabase.anonKey = 'eyJ...your-anon-key';

Start the backend

cd DualMind_Back/src/DualMind.API
dotnet restore
dotnet run

Verify: curl http://localhost:5079/health returns {"status":"healthy"}

Start the frontend

cd DualMind_UI
npm install
npm run dev

Verify: Open http://localhost:8000 in your browser

Your first arena battle

Via the UI
Via cURL

Open http://localhost:8000
Click Login with Google (or create an account via Supabase)
Type a prompt like “Explain quantum computing simply”
Click Compare — two models respond side-by-side
Vote for the better response
Model names are revealed after voting

Get a JWT token by logging in through the frontend, then:

curl -X POST http://localhost:5079/api/arena/chat \
  -H "Authorization: Bearer YOUR_JWT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"prompt": "Explain quantum computing in simple terms"}'

{
  "object": "ai.response",
  "success": true,
  "message": "Quantum computing is a type of computing that...",
  "model": {
    "name": "llama-3.3-70b-versatile",
    "displayName": "Llama 3.3 70B",
    "provider": "groq"
  },
  "responseTimeMs": 1245,
  "usage": { "promptTokens": 15, "completionTokens": 120, "totalTokens": 135 }
}

curl -X POST http://localhost:5079/api/arena/dualchat \
  -H "Authorization: Bearer YOUR_JWT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"prompt": "Explain quantum computing", "selectionMode": "random"}'

The response includes agent1, agent2, comparisonId, and arena comparison metrics.

Copy the comparisonId from the dual-chat response to submit a vote via POST /api/arena/model-vote.

You now have a running DualMind Lab instance with chat, comparisons, and voting.

Next steps

Architecture Overview

Understand the full system design

API Reference

Explore all API endpoints

Database Schema

See all tables and relationships

Troubleshooting

Fix common setup issues

Common issues

401 Unauthorized

Root cause: JWT token is missing, expired, or signed with wrong secret.Fix:

Verify JWT token is from Supabase Auth (not expired)
Check Authorization: Bearer <token> header format
Ensure JWT_SECRET in .env matches your Supabase project

# Paste your token at jwt.io to inspect claims
# Verify: iss matches SUPABASE_URL/auth/v1, aud is "authenticated"

500 on chat requests

Root cause: AI provider failure or missing database records.Fix:

Check GROQ_API_KEY is valid at console.groq.com
Ensure ai_models table has at least one row with status = 'active'
Check backend terminal logs for the specific provider error

The fallback chain tries: Groq primary → Groq alt model → Bytez. All three must fail for a 500.

Frontend can't reach backend

Root cause: Backend not running or CORS misconfiguration.Fix:

Verify backend is running: curl http://localhost:5079/health
Check config.js — apiBaseUrl auto-detects localhost in development
Open browser DevTools > Console for specific CORS or network errors

In production, the Cloudflare Worker proxies /api/* to the backend, so CORS is not an issue.

Getting Started

Core Concepts

Architecture

Database

Frontend Guide

Development

Deployment

Prerequisites

Setup

Next steps

Architecture Overview

API Reference

Database Schema

Troubleshooting

Common issues

​Prerequisites

​Setup

​Next steps

Architecture Overview

API Reference

Database Schema

Troubleshooting

​Common issues

Prerequisites

Setup

Next steps

Common issues