Concordia
Generative social simulation with LLM agents
About
Concordia is Google DeepMind's library for building generative agent-based models that simulate interactions among entities in physical, social, or digital environments. It uses a tabletop-roleplaying-inspired pattern: a Game Master entity simulates the environment while player agents describe intended actions in natural language, and the GM turns those into outcomes. Use cases span social science research, AI safety and ethics, economics, synthetic data generation, and service evaluation through simulated usage.
Version 2.0 rebuilt the library on an Entity-Component pattern with prefabs for reusable scenarios, serialization, and checkpointing. It installs via pip install gdm-concordia, needs an LLM API plus a text embedder for associative memory, and is documented in two arXiv tech reports (2312.03664 and 2507.08892).
What it can do
- Game Master environment simulation
- Natural-language agent actions
- Entity-Component architecture (v2)
- Prefabs for reusable scenario building
- Serialization and checkpointing
- Associative memory with embeddings
Setup
- Install with pip install gdm-concordia
- Get an LLM API key and a text embedder for associative memory
- Define entities (agents) and a Game Master with components
- Run the episode loop and inspect event statements
Prerequisites
Python 3.10+, access to an LLM API, and a sentence embedding model.
Use cases
Works with
Trust score — 8/10
Scored 8/10 (TIER verified, metrics-v1). Evidence: 1,789 GitHub stars (api.github.com, 2026-10-16); pushed 2026-10-15; Apache-2.0; Google DeepMind authorship with two arXiv tech reports; used in the NeurIPS 2024 Concordia Contest. Strengths: strong research pedigree, active maintenance, peer-reviewed design. Docked for: (1) simulation fidelity depends heavily on the chosen LLM; (2) niche audience outside research and evaluation teams. License: Apache-2.0.
★ 1,789 GitHub stars
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.0/10
- Maintenance · recency, releases, issue load
- 9.0/10
- Openness · license + docs
- 9.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · 1,789 GitHub stars via api.github.com/repos/google-deepmind/concordia (2026-10-16)
- · Last push 2026-10-15T17:37:38Z - actively maintained
- · Two arXiv tech reports: 2312.03664 and 2507.08892
- · Used in the NeurIPS 2024 Concordia Contest (cooperativeai.com)
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.