AI Agent UN: simulating the General Assembly with AI diplomats

I've been fascinated by the idea of using AI agents to simulate complex multi-party interactions, and international diplomacy is about as complex as it gets. So I built an AI-powered Model United Nations where each agent embodies a specific country's foreign policy positions, diplomatic style, and national interests.

How it works

AI Agent UN works by giving each country a detailed system prompt that defines their foreign policy framework. When you submit a motion — say, a Gaza ceasefire resolution — each agent reads the motion text and responds with a structured JSON vote (yes, no, or abstain) along with a statement explaining their position. The system collects all the votes and compiles the results.

You can run simulations with the full membership or sample just a handful of countries for quick testing. It supports both cloud APIs (OpenAI, Anthropic) and local models via Ollama, so you can run the whole thing without any API costs if you prefer.

Beyond simple voting

What makes this more interesting than a simple poll is the analysis layer. The repo includes tools for bilateral impact analysis — you can see how a particular vote pattern might strengthen or strain relationships between specific country pairs. There's also CSV export and PDF report generation for deeper analysis. The results end up being surprisingly nuanced, with agents articulating positions that track reasonably well with real-world diplomatic stances.

Important caveats

I want to be clear: this is an experimental and educational tool, not a prediction engine. AI agent positions don't represent actual government policies. The value lies in the modeling approach and system architecture — exploring how multi-agent systems handle complex scenarios with many competing interests. It's a proof of concept for using agentic AI in geopolitical simulation, and I think there's a lot of untapped potential in this space for education, research, and scenario planning.