Debate GPT

A local web app where two language models debate a question with separate private contexts.

Features

• Positive and negative debater personality prompts
• Separate model selection for each debater
• Separate private context for each debater
• Public transcript visible to both agents
• Private thought side panels
• Thought mode and thought amount controls
• Clickthrough or uninterrupted debate
• User interjections that both debaters see on future turns
• Session-only state

Quick Start

Install dependencies:

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

If python3-venv is not available, install dependencies into the repo-local .deps directory instead:

python3 -m pip install -r requirements.txt -t .deps

Install and run Ollama, then pull a model:

ollama pull llama3.2
ollama serve

Start Debate GPT:

python3 server.py

Open:

http://127.0.0.1:8000

Local Model Endpoint

The default endpoint is:

http://localhost:11434/v1

That matches Ollama's OpenAI-compatible API. You can also point the UI at another local OpenAI-compatible server, such as LM Studio or llama.cpp server.

Environment defaults:

LOCAL_MODEL_BASE_URL=http://localhost:11434/v1
LOCAL_MODEL_NAME=llama3.2
PORT=8000

Notes

The backend asks each model turn to return JSON:

{
  "thought": "private thought shown in that debater's side panel",
  "response": "public message shown in the debate transcript"
}

The other debater only receives the shared public transcript and user interjections, never the opponent's private thought.