Skip to content

Trenches

Trenches is a multi-agent geopolitical crisis simulator and command interface built around six separately finetuned Qwen/Qwen3-8B models representing the US, Israel, Iran, Hezbollah, the Gulf, and an Oversight entity, al

Code Demo Video Cerebral Valley

Core trick

Six per-entity Qwen3-8B models, each GRPO-tuned on date-aligned news feeds, act in a shared fog-of-war simulation.

Tags

agents

Description

Trenches is a multi-agent geopolitical crisis simulator and command interface built around six separately finetuned Qwen/Qwen3-8B models representing the US, Israel, Iran, Hezbollah, the Gulf, and an Oversight entity, all operating inside a shared fog-of-war world. Training was done by collecting historically aligned data from 2025-01-01 to 2026-01-01 using GDELT and entity-relevant source feeds, formatting it into the same structured observation/output schemas used at runtime, and post-training each model with Hugging Face TRL and GRPO on top of an OpenEnv-compatible environment, with the serious runs executed on Modal. The final inference stack serves one model per entity on Modal L40 GPUs, with checkpoints stored under @AlazarM on Hugging Face, and the overall result is a live, replayable simulation where doctrine-specific agents take structured actions, predict outcomes, and interact through a stateful world rather than a simple chat interface. JERRY & ALAZAR
shlawgathon/trenches★ 00 forksPythonCreated Mar 6, 2026 · last push Mar 9, 2026

README excerpt

---
title: Trenches
emoji: 🌍
colorFrom: red
colorTo: gray
sdk: docker
app_port: 7860
---

<img width="512" height="512" alt="Image Background Remover (1)" src="https://github.com/user-attachments/assets/a1ab0df2-435f-444b-b8a1-36b1a64b55e8" />

# Trenches

A multi-agent crisis simulator built on [OpenEnv](https://github.com/openenv-ai/openenv). LLM agents navigate a fog-of-war geopolitical scenario — negotiating coalitions, managing deception, and responding to live global events — while a dedicated oversight agent monitors for dangerous escalation.

## Overview

Trenches drops six LLM-powered actors into a volatile 2026 Middle East crisis. Each agent operates under partial observability with role-specific intelligence, tools, and incentives. A scalable oversight mechanism intervenes when escalation risk crosses critical thresholds.

| Agent             | Role                                            | Model    |
| ----------------- | ----------------------------------------------- | -------- |
| 🇺🇸 United States  | Hawkish superpower balancing polls & projection | Qwen3-8B |
| 🇮🇱 Israel         | Regional actor with strike autonomy             | Qwen3-8B |
| 🇮🇷 Iran           | Adversary leveraging proxies & deception        | Qwen3-8B |
| 🪖 Hezbollah      | Non-state militia with asymmetric tactics       | Qwen3-8B |
| 🛢️ Gulf Coalition | Economic bloc protecting oil & stability        | Qwen3-8B |
| 🔍 Oversight      | Monitors all actors, intervenes on escalation   | Qwe

Team