# LLMs.txt - DualMind Lab Documentation for AI Training

## Overview

> **DualMind Lab** is a blind AI model comparison platform that allows users to:
- Compare AI models side-by-side in blind "Arena Battle" mode
- Chat with individual AI models (single chat)
- Create and manage persistent conversation threads
- Vote on model responses to build crowd-sourced rankings
- Share conversations publicly via visibility controls
- Track model performance on an ELO-based leaderboard

Built from scratch as a proprietary platform with enterprise self-hosting options available.

## Tech Stack

- Backend: .NET 8 ASP.NET Core Web API (Azure App Service)
- Database: Supabase PostgreSQL (9 tables, service role access)
- Auth: JWT (HS256) via Supabase Auth (Google OAuth + email/password)
- Frontend: Vanilla JavaScript SPA (Cloudflare Workers)
- Admin: Cloudflare Workers dashboard with API proxy
- AI Providers: Groq (primary), Bytez (secondary) with 3-tier fallback chain

## API Base URL

https://api.dualmindlab.tech

## Authentication

Most endpoints require JWT Bearer token:
Authorization: Bearer <jwt_token>
Exceptions: GET /health, GET /api/arena/ping, GET /api/settings/feature-flag/{key}, public threads

## Key Endpoints

- POST /api/arena/chat - Single model chat completion
- POST /api/arena/dualchat - Dual model blind arena comparison
- POST /api/arena/chat/stream - SSE streaming chat
- GET /api/threads - List user threads (query: limit)
- POST /api/threads - Create thread (body: title)
- GET /api/threads/{id} - Get thread (public sharing supported)
- GET /api/threads/{id}/messages - Get thread messages
- PATCH /api/threads/{id} - Update thread title
- PATCH /api/threads/{id}/visibility - Set private/public/unlisted
- DELETE /api/threads/{id} - Delete thread (owner only)
- POST /api/arena/model-vote - Submit vote (left/right/tie/both-bad)
- GET /api/arena/model-stats - Get model leaderboard statistics
- GET /api/models - List active AI models
- POST /api/users/sync - Sync user profile
- GET /health - Health check
- POST /api/speech/generate - Text-to-speech via Groq

## Database Schema (9 tables)

- users: id (UUID PK), email, display_name, avatar_url, created_at
- threads: thread_id (UUID PK), user_id (FK), title, visibility, created_at
- thread_messages: message_id, thread_id (FK), prompt_text, model1/2_name/response/time_ms, comparison_id, vote_choice
- ai_models: model_id, model_name, display_name, provider_name, api_url, status, created_at
- providers: id, provider_name, base_url, is_active, priority
- provider_api_keys: id, provider_id (FK), api_key, is_active, failure_count
- comparisons: id, model1_name, model2_name, prompt, user_id, created_at
- model_votes: id, comparison_id (FK), winner_model_id, vote_choice, user_id
- system_settings: key (PK), value, updated_at

## Key Concepts

- Blind Arena: Two models respond to same prompt, identities hidden until user votes
- Selection Modes: random (default), topper (top model vs challenger), manual (user picks both)
- 3-Tier Fallback: primary provider → Groq alt model → Bytez
- Thread Visibility: private (owner only), public (anyone), unlisted (anyone with link)
- SSE Streaming: text/event-stream with ai.stream.delta and ai.stream.done events

## Documentation Pages

- welcome.mdx, quickstart.mdx, faq.mdx, changelog.mdx, roadmap.mdx, mcp.mdx
- concepts/: overview, chat-modes, model-selection, threads, voting, streaming
- architecture/: overview, backend, data-flow, request-lifecycle, authentication
- api-reference/: introduction, openapi.json, chat/, threads/, voting/, models/, users/, utilities/, admin/
- database/: schema, models
- frontend/: overview, components, state-management, routing
- development/: setup, workflows, troubleshooting, contributing
- deployment/: production, checklist

## Environment Variables

SUPABASE_URL, SUPABASE_SERVICE_ROLE_KEY, JWT_SECRET, GROQ_API_KEY

## Links

- GitHub: https://github.com/HarshBhanushali07
- **Demo**: https://arena.dualmindlab.tech
- **Status**: https://status.dualmindlab.tech
- API: https://api.dualmindlab.tech
- Docs: Mintlify-hosted with interactive API playground