LangChain + Ollama Autonomous debate Markdown output

Autonomous, turn-based debates between two LLM agents.

PRO vs CON with periodic analysis checkpoints and dynamic restrictions to prevent loops and keep long debates coherent.

See examples

Overview

Two debaters

PRO and CON agents debate a topic over many turns with Ollama-compatible models.

Anti-loop checkpoints

A summarizer analyzes the transcript periodically and injects restrictions to cut repetition and stalemates.

Judge

Can end the debate early on agreement, total refutation, or clear concession.

How it works

1

Turn-based debate

PRO and CON alternate responses with controlled conversation memory.

2

Periodic analysis

The summarizer extracts arguments, refutations, and exhausted topics.

3

Restriction injection

Prompts are updated to avoid repetitive lines and force novelty.

4

Judge

Can end the debate when there is agreement or a clear winner.

Real examples

Real CLI outputs available in the docs/ folder.

CLI transcript Source: 225.txt 
Loading…

Tip: you can open the raw file too: 225.txt