A Society of Agents

Inspired by Marvin Minsky's Society of Mind, a multi-agent simulation where AI personalities debate questions from different perspectives. Ask a question, specify the viewpoints you want, and watch the debate unfold.

What it does

1. You provide a question and the perspectives you want represented
2. AI generates distinct agent personalities for each perspective
3. An orchestrator facilitates debate between the agents
4. Agents discuss, disagree, and build on each other's arguments

Installation

pip install pysociety

or

uv add pysociety

Quick Start

Create a .env file with your API key:

ANTHROPIC_API_KEY=your-api-key-here

Run a debate

uv run python -m llm_society \
  "What is the meaning of life? I want an optimistic, a pessimistic, and a pragmatic view."

More examples

# Ethics debate
uv run python -m llm_society "$(cat examples/trolley_problem.txt)"

# Philosophy
uv run python -m llm_society "$(cat examples/free_will.txt)"

# Social issues
uv run python -m llm_society "$(cat examples/wealth_inequality.txt)"

Example Questions

Optional Configuration

You can customize models in .env:

# Main orchestrator model
MAIN_AGENT_MODEL=anthropic:claude-opus-4-5-20251101

# Sub-agent models
SUB_AGENT_MODEL=anthropic:claude-sonnet-4-5-20250929

# Prompt generation model
PROMPT_GENERATOR_MODEL=anthropic/claude-opus-4-5-20251101