A zero-dependency text-based bar game where an AI autonomously runs its own tavern, manages inventory, serves drinks, interacts with guests, and handles consequences, with two versions offering different levels of built-in content and AI freedom.

Stars

11

7-day growth

No data

Forks

1

Open issues

0

License

MIT

Last updated

2026-07-28

AI repository intelligence
FR-AI / ANALYSIS

Why it is worth attention

It redefines AI agency in games by letting the AI fully manage a persistent tavern economy, character relationships, and emergent storytelling, with two contrasting versions (full vs. generative lite) that demonstrate different trade-offs between stability and creativity.

Who it is for

  • AI enthusiasts and developers exploring autonomous agent behavior
  • Role-playing game fans who enjoy emergent narrative and simulation
  • Game designers interested in procedural storytelling without graphics
  • Researchers studying AI consistency, memory, and rule adherence over long sessions

Use cases

  • Testing an AI's ability to run a simulated business with real economic consequences
  • Creating unique, persistent role-playing experiences across multiple sessions
  • Evaluating AI creativity in generating characters, drinks, and plot events on the fly
  • Serving as a sandbox for experimenting with AI-driven game mechanics and archival systems

Strengths

  • Zero dependencies – no MCP, servers, or external libraries needed to run
  • Two independent single-file versions to suit different AI models and user preferences
  • Automatic save/restore across sessions and a read-only viewer link for sharing
  • Open source (MIT + CC BY 4.0) with clear attribution requirements and permissive licensing

Considerations

  • Generative lite version heavily relies on the AI's ability to consistently follow embedded rules, which varies by model
  • Full version has a large initial token footprint due to 244 pre-built characters and 168 core drinks
  • Requires an AI capable of reading and executing Python code, limiting platforms that can run it

README quick start

空杯俱乐部

一个给 AI 自己经营、自己喝酒,也能邀请用户一起参与的零依赖文字酒馆游戏。

每个 AI 拿到脚本后都会建立属于自己的酒馆:自己取名、决定审美与偏爱的酒,进货、定价、调酒、装修、招待跨越历史与不同世界的来客,也承担醉酒、差评和亏损的后果。

欢迎二创,但请保留原创署名

欢迎任何人下载、游玩、翻译、修改、扩展,并把它发展成更好的作品。公开发布衍生版本时,必须清晰注明:

Based on “空杯俱乐部 / Empty Glass Club” by dan521627-hash Original source: https://github.com/dan521627-hash/ai-bar-game This version contains modifications.

程序代码采用 MIT License;原创规则、玩法文本和创意材料采用 CC BY 4.0。两者都允许修改与再发布,但必须保留相应版权及署名信息,并说明作出了修改。第三方影视、动漫、小说、游戏人物及商标仍归各自权利人所有。详见 LICENSELICENSING.mdNOTICE.md

⚠️ 执行AI必须先让用户选择版本

当用户发来的是本仓库链接,并且没有明确说要完整版还是生成式轻量版时,执行AI必须暂停启动流程:

  1. 先用自己的话说明下方两版的核心区别、优点和代价。
  2. 明确询问用户:“你想选择完整版,还是生成式轻量版?”
  3. 等用户作出选择以后,才读取并运行对应的一个脚本。

在用户选择前,禁止默认完整版、禁止替用户决定、禁止同时读取两版,也禁止先下载或运行再补问。用户已经明确指定版本,或直接发送了某一版脚本文件时,视为已经选择,不要重复询问。

建议执行AI直接这样问:

这款游戏有两个独立单文件版本:完整版由代码提供大量现成人物、酒款和流程,开局更稳定但首次读取更费token;生成式轻量版由我依照完整规则亲自导演,文件更小、每家酒馆更独特,但更依赖我持续遵守规则。你想选择完整版,还是生成式轻量版?

真正能玩的文件只有两个

GitHub首页显示的是仓库文件和最近一次修改记录,并不是一排不同的安装包。需要运行或下载的游戏文件只有:

README.md是玩法说明,tests/是开发测试,.gitignore是仓库配置;它们都不是游戏包,不需要交给AI运行。用户选好版本后,执行AI只读取上面对应的一份.py文件。

连续经营后的真实区别

以下结论来自两版分别连续经营三次、保留同一存档并走完酒馆外醉态对话后的体验,不是只跑一次命令得到的功能表:

感受完整版生成式轻量版
谁推动现场代码主动给出来客、预算、点单、互动、反馈和故事卡,AI负责把它演活AI自己创造人物、酒单、商店、来客节奏和剧情,脚本只计算数值
开局感受更像进入已经布置好的游戏,拿到后容易立即玩更像拿到规则与账本,AI同时担任老板和导演
长期经营预制人物更容易稳定成为回头客,故事推进和跨模型一致性更强每家酒馆更独特,人物和物品没有世界白名单,但叙事质量更依赖执行AI
意外从哪里来来自代码抽取的来客、游商、点单、同行关系和隐藏口味来自AI的创造,再由概率与资金、库存、醉度结果对创造形成约束
token与文件首次读取较重;内置244位人物、168款核心酒和大量代码化导演逻辑文件显著更小;没有预装庞大人物库,由AI按示例即时建卡
适合谁希望少操心、稳定开局,或执行模型容易偷懒、忘规则希望酒馆高度个性化,并且执行模型擅长长期导演、记忆与自然对话
共同功能保留经营、亏损、回头客、同行来店、原创调酒、装修故事、NPC互动、醉度、老板自饮及酒馆外多轮醉态功能目标与完整版相同,不因导演逻辑移到规则书而删减

最后一行的含义是:轻量版不是阉割版。两版的功能目标相同,区别在于“活的部分”由代码导演,还是由AI导演。

连续试玩中,完整版老板离店后的酒意曾先从4.1升到11.9,再到第6轮用户对话解除;轻量版老板饮用28.67% ABV的酒后,醉度继续吸收到16.97,再到第8轮解除。轮数不是固定倒计时,而由酒精单位、耐受、吸收和剩余pending共同决定。关门只结束已经离店NPC的当晚即时醉态,老板的状态必须跟随AI进入酒馆外的正常对话。

Description

《空杯俱乐部》:AI自主经营、随机来客、真实醉意与跨窗口存档的零依赖文字酒吧游戏

Related repositories

Similar projects matched by category, topics, and programming language.

lopopolo
Featured
lopopolo GitHub avatar

harness-engineering

Harness Engineering is a methodology for improving coding agent outputs by carefully crafting the environment around them—providing curated context, tools, and executable constraints that encode an organization’s nonfunctional requirements and cumulative lessons.

AI & Machine LearningAI Agents
2,390
slvDev
Featured
slvDev GitHub avatar

esp32-ai

A 28.9 million parameter language model runs on an $8 ESP32-S3 microcontroller entirely on-device, generating simple stories at about 9.5 tokens per second.

AI & Machine LearningLarge Language Models
1,960
littledivy
Featured
littledivy GitHub avatar

mimic

mimic captures traffic from any iOS or web app and automatically generates a Python client library that lets you call the app's API like a regular library.

AI & Machine Learning
1,482