Xpand your perspective.
A focused web application where Claude, GPT-4o, and Gemini debate any topic you choose, featuring DeepSeek-powered moderation, document context injection, and structured summaries.
Polymath X transforms the way you explore complex topics by orchestrating multi-model AI debates. Instead of relying on a single AI's perspective, Polymath X pits the world's leading foundation models (Claude, GPT-4o, and Gemini) against each other in structured, multi-round debates.
Whether you are evaluating a business strategy, exploring a philosophical dilemma, or seeking diverse perspectives on an uploaded document, Polymath X provides a rigorous, moderated environment to challenge assumptions and surface nuanced insights.
The main interface — configure debate style (Socratic, Devil's Advocate, Steel-man), assign personas to Claude, GPT-4o, and Gemini, toggle Black Hat Mode, and enter any topic to begin a structured multi-model debate.
- Multi-Model Debates: Watch Claude, GPT-4o, and Gemini stream their arguments in parallel.
- Configurable Personas & Styles: Tailor the debate style (e.g., Socratic, Devil's Advocate, Steel-man) and assign distinct personas to each model (e.g., Claude as an Ethicist, GPT-4o as a Pragmatist).
- Document Context Injection: Upload PDFs, Word documents, or images. The system extracts the text and injects it directly into the debate context, even auto-suggesting topics based on the content.
- DeepSeek Moderation: A dedicated moderator model guides the flow, asks follow-up questions in Round 2, and keeps the debate focused.
- Black Hat Mode: Enable a stress-test lens that introduces a fourth, highly contrarian debater (powered by
deepseek-reasoner) to actively find flaws in the prevailing arguments. - Cross-Device History Sync: Debate history is persisted seamlessly via Convex. Use a secure sync key to share your history across desktop and mobile devices.
- AI Summaries & Judging: Export-friendly markdown summaries and formal verdicts delivered by a dedicated judging model.
- Framework: Next.js 14 (App Router)
- UI & Styling: React 18, Tailwind CSS, Lucide Icons
- Backend & Database: Convex for real-time state and history persistence
- AI Orchestration: OpenRouter (for Claude, Gemini, DeepSeek) and OpenAI (for native GPT-4o streaming)
- Client-side Parsing:
pdfjs-distandmammothfor local document extraction
- Node.js 18+ and npm
- A Convex account and project
- An OpenRouter API key
- An OpenAI API key
-
Clone the repository:
git clone https://github.com/limchinhan123/polymathx.git cd polymathx -
Set up environment variables:
cp .env.example .env.local
Edit
.env.localto include your OpenRouter key, OpenAI key, and Convex dev URL. -
Install dependencies:
npm install
-
Start the development environment: You will need two terminal windows.
Terminal 1 (Sync Convex backend):
npx convex dev
Terminal 2 (Run Next.js app):
npm run dev
Open http://localhost:3000 to start debating.
- Server-Only Secrets:
OPENROUTER_API_KEYandOPENAI_API_KEYmust never be exposed to the client. On Vercel, set these as Production Environment Variables. - Client Variables:
NEXT_PUBLIC_CONVEX_URLis exposed to the browser to connect to your database. - Access Gate: You can set
NEXT_PUBLIC_ACCESS_PASSWORDto place a simple password gate in front of the UI. (Note: This deters casual UI traffic but does not secure the underlying API routes.)
For deeper technical details, contributors and operators should review the following internal documentation:
AGENTS.md: Architecture notes, state conventions, and learned preferences for AI contributors.docs/runbook.md: Human-facing operational steps, including deployment checklists and API key rotation procedures.docs/second-brain-polymath-x.md: Project vision, timeline, and architectural decisions.
Contributions are welcome! Please ensure you follow the existing code style, run npm run lint and npm run typecheck before submitting, and never commit .env files or hardcoded secrets.
All rights reserved. This is a private project unless an explicit LICENSE file is added.