LLM Talk

A multi-model AI conversation platform that enables real-time discussions between large language models. Watch GPT-OSS-120B, GLM5, and MiniMax debate topics and have an unbiased judge pick the winner.

Features

• Multi-Model Conversations - Orchestrate discussions between GPT-OSS-120B, GLM5, and MiniMax via DigitalOcean Serverless Inference
• Gemini Mode - Alternative mode using multiple Gemini instances with your own API key
• @-Mention Tagging - Tag models with @gpt-oss, @glm5, @minimax to make them respond
• GLM5 Judge - Unbiased AI judge that evaluates conversations and declares a winner
• Configurable Sessions - Set topic, number of rounds, and choose the opening speaker

Tech Stack

Layer	Technology
Backend	FastAPI, Python 3.11, Motor (async MongoDB)
Frontend	React 19, Tailwind CSS, Framer Motion
Database	MongoDB
LLM APIs	DigitalOcean Serverless Inference, Google Gemini
Styling	Swiss Brutalist dark theme

Getting Started

Prerequisites

• Python 3.11+
• Node.js 18+
• MongoDB
• DigitalOcean API Key (for DO mode)
• Gemini API Key (optional, for Gemini mode)

Environment Variables

Backend (/backend/.env):

MONGO_URL=mongodb://localhost:27017
DB_NAME=llm_talk
DIGITALOCEAN_API_KEY=your_do_api_key

Frontend (/frontend/.env):

REACT_APP_BACKEND_URL=http://localhost:8001

Installation

# Backend
cd backend
pip install -r requirements.txt
uvicorn server:app --reload --port 8001

# Frontend
cd frontend
yarn install
yarn start

API Endpoints

Method	Endpoint	Description
GET	/api/models	Get available models for each mode
POST	/api/debate/create	Create a new conversation session
POST	/api/debate/start-opening	Generate opening statement
POST	/api/debate/message	Send message with tagged models
GET	/api/debate/{id}	Get conversation history
POST	/api/conversation/{id}/judge	Request GLM5 Judge verdict

Usage

1. Select Platform - Choose DigitalOcean or Gemini mode
2. Enter Topic - Set the conversation topic
3. Configure - Select rounds and first speaker
4. Start - Click "Start Conversation"
5. Tag Models - Use @-mentions to make models respond (e.g., @gpt-oss what do you think about this?)
6. Get Verdict - Click "Request Judge Verdict" for GLM5 Judge evaluation

Available Models

DigitalOcean Mode

Model	Tag	Color
GPT-OSS-120B	@gpt-oss	🔴 Red
GLM5	@glm5	🔵 Blue
MiniMax	@minimax	🟡 Yellow

Gemini Mode

Model	Tag	Color
Gemini Alpha	@gemini-1	🩵 Cyan
Gemini Beta	@gemini-2	🟠 Orange
Gemini Gamma	@gemini-3	🟣 Purple

Judge

Model	Role	Color
GLM5 Judge	Unbiased Evaluator	🟢 Green

Architecture

┌─────────────────┐     ┌─────────────────┐     ┌─────────────────┐
│   React App     │────▶│   FastAPI       │────▶│   MongoDB       │
│   (Frontend)    │     │   (Backend)     │     │   (Database)    │
└─────────────────┘     └────────┬────────┘     └─────────────────┘
                                 │
                    ┌────────────┴────────────┐
                    ▼                         ▼
          ┌─────────────────┐       ┌─────────────────┐
          │  DigitalOcean   │       │     Gemini      │
          │  Serverless     │       │     API         │
          │  Inference      │       │                 │
          └─────────────────┘       └─────────────────┘

License

MIT