CivAgent
Enrichment pendingCivAgent is an LLM-based Human-like Agent acting as a Digital Player within the Strategy Game Unciv.
GraphCanon updated today · GitHub synced today
Trust & integrity
Full report- Maintenance
- Dormant (481d since push)
- As of today · Source: github_public_v1
- Provenance
- Not a fork · Organization account
- As of today · Source: github_public_v1
- Security (OSV)
- 43 low (43 low)
- As of today · Source: osv@v1
Public GitHub metadata and optional OSV dependency scans. Signals, not a guarantee. Trust methodology.
Overview
CivAgent is an LLM-based Human-like Agent acting as a Digital Player within the Strategy Game Unciv.
Capability facts
- Languages
- python
Source: github.language+pyproject.toml · Jul 11, 2026
Categories
Tags
README
CivAgent: Large Language Models based Human-like Agent
English | 中文 |
Moved the project repository to a new location: https://github.com/asdqsczser/CivAgent.
CivAgent is an LLM-based Human-like Agent acting as a Digital Player within the strategy game Unciv. The project aims to address the difficulty of players not being able to find human playmates, and seeks to construct a low-cost data flywheel to aid in the research of LLM-based Agents. By deeply integrating Large Language Models (LLMs) with core gameplay, we believe this is exactly an AI Native Game. All glory are attributed to Unciv.
Native Unciv: https://github.com/yairm210/Unciv
Unciv Wiki: https://civilization.fandom.com/wiki/Unciv
Game Client Download (developer version, no Discord chat function): Mac Windows
Game Client Download (player version, with Discord chat function, to be released in early August)
Unciv We Used (source code): https://github.com/asdqsczser/Unciv/tree/release/fuxi_ver
Paper: https://arxiv.org/pdf/2502.20807
Appendix: https://github.com/fuxiAIlab/CivAgent/blob/main/paper_appendix.pdf
YouTube: https://www.youtube.com/playlist?list=PL9G00-od8ezYNblsqMca7urrLoHSZ7Vs3
News
03/17/2025: Moved the project repository to a new location: https://github.com/asdqsczser/CivAgent.
06/20/2024: Code for benchmark reproduction (developer version) of research paper is open-sourced. The version for players is expected to be released in early July.
06/09/2024: The paper is submitted to NeurIPS 2024 Track Datasets and Benchmarks, Under Review.
👨💻 Developers
Please refer to Documentation.
🙋 Contribution
Please refer to Documentation.
⚡ How to play
The AI service can only be used in the multiplayer game mode and is only supported on the Windows and Mac platform.
Note that your save files and chat data will be collected by the server (no personally identifiable data) and used for non-commercial purposes (AI effect improvement for this project).
Step 1. Download the game client
Released in early July.
Step 2. Set the server address
Open the game, navigate to the 'Options' and 'Multiplayer' interface, adjust the synchronization frequency to 3 seconds, modify the server address to http://sl.office.fuxi.netease.com:44952 and test the link.
Step 3. Start a new game
Open the 'Start new game' interface, check the 'Online Multiplayer' option in the bottom left corner. It is recommended to begin with the 'Medieval era'. On the right, please click on 'Set current user', and select a civilization from the following civilizations: China, Mongolia, Egypt, Aztecs, Rome, or Greece.
Step 4. Copy the GameId
After creating the game, the GameId will be automatically copied to your clipboard, or you can find the GameId of the game you created in the multiplayer game interface.
Step 5. Chat with civilizations on the Discord
Step 6 (optional). Enhancing AI experience using LLMs such as GPT-4
The current default AI service utilizes the free large-scale model. For an enhanced gaming experience, you may consider subscribing to OPENAI's GPT-4 service.
🙋 Motivation & Discussion
To Fans of Strategy Games
In strategy games, the experience of multiplayer gameplay is much better than that of computer AI, primarily due to the fact that human players make more rational strategic decisions, have a flexible diplomatic strategy, and can employ natural language for negotiation and deception. A typical example is when the emergence of a new conq