Concordia is a Python library for building generative agent-based simulations, targeting researchers and developers in social sciences, AI ethics, and economics. It simplifies the creation of complex agent interactions within simulated environments by employing a "Game Master" (GM) agent that interprets natural language actions and orchestrates environmental responses, enabling flexible integration with LLMs and external services.

How It Works

Concordia models agent interactions through a Game Master (GM) paradigm, inspired by tabletop role-playing games. Agents describe their desired actions in natural language, and the GM translates these into simulated environmental effects, handling physical plausibility or API calls to external digital services. This approach allows for rich, emergent behaviors driven by LLM-powered agent reasoning and a flexible, extensible simulation architecture.

Quick Start & Requirements

• Install via pip: pip install gdm-concordia
• Requires access to a text-generating LLM API and a text embedder for associative memory.
• Editable install: git clone https://github.com/google-deepmind/concordia && cd concordia && pip install --editable .[dev]
• Development environment available via devcontainer.

Highlighted Details

• Facilitates agent reasoning based on the "What kind of situation is this? What kind of person am I? What does a person such as I do in a situation such as this?" framework.
• Supports simulations grounded in physical, social, or digital spaces.
• Can be used for data generation for personalization and performance evaluation of real services.
• Integrates with external applications and services via the GM.

Maintenance & Community

• Developed by Google DeepMind.
• The project is presented alongside a technical report and research paper.

Licensing & Compatibility

• License: Apache License 2.0.
• Compatible with commercial use and closed-source linking.

Limitations & Caveats

The library is presented as "not an officially supported Google product." The quality of simulations is dependent on the chosen LLM and embedder.