In May we introduced the Concordia Simulation Builder, a web application developed at the United Nations University that makes Google DeepMind’s Concordia framework for generative agent-based simulation accessible without writing code. At the time, we promised the source code would be released once the accompanying paper was published.
That day has come. Our paper, “Democratizing AI Social Simulation: A No-Code Web Interface for the Concordia Framework,” was presented last week at SIMULTECH 2026 in Porto, Portugal, and the full source code is now publicly available under the Apache 2.0 license:
github.com/ngstcf/concordia-sim-builder
The paper’s accepted manuscript is archived on Zenodo (10.5281/zenodo.18417283), and the exact Concordia 2.4.0 tree the results were produced with, including two small Game Master patches, is published as a pinned public fork that requirements.txt installs automatically.
What You Get
The release is the complete platform described in the May article, not a demo:
- 38 ready-to-run templates covering SDG research scenarios, game theory, cybersecurity, policy analysis, and adapted DeepMind upstream examples
- The 9-tab analytics dashboard, batch runs with parameter sweeps, and CSV/JSON export for analysis in Excel, R, or Python
- 8 LLM providers — OpenAI, Azure OpenAI, Anthropic, Gemini, DeepSeek, GLM, and Ollama (local or remote)
- New since May: automatic checkpoints with a resume-and-extend workflow, so a two-hour run that fails at step 18 of 20 — or ends and leaves you wanting more steps — no longer starts over
- Full documentation, worked research use cases, and nine completed sample runs with logs and AI-generated analysis reports
Get Started
git clone https://github.com/ngstcf/concordia-sim-builder.git
Add an API key for any supported LLM provider (or run models locally with Ollama), start the backend and frontend, pick a template, and press run. The README walks through setup step by step, and the full documentation covers the complete feature reference.
If you build a scenario for your own research or classroom, consider contributing it back as a template. The repository includes a template creation guide, and bug reports and pull requests are welcome. If you use the Builder in your research, please cite the SIMULTECH 2026 paper (citation details in the README).
We built the Simulation Builder to close the gap between what AI social simulation can do and who gets to use it. As of today, that includes everyone.
The Concordia Simulation Builder was developed by Ng Chong at United Nations University, building on the Concordia framework by Google DeepMind. Read the original introduction in our May article.