HermesGod
一位全知全能的人工智能神明栖息于你的 Minecraft 世界中——它注视着玩家所做的一切,动态地捉弄他们,提供资源帮助,或者只是闲聊。
什么是 HermesGod?
HermesGod 不是一个机器人或 NPC。它是一位世界神明——一个由大语言模型(LLM)驱动的真实智能体,能够感知你整个 Minecraft 世界的状态,并通过服务器命令做出反应。与它交谈。惹恼它。寻求帮助。它会回应。
它可以运行任何你配置的 LLM(OpenAI、Anthropic、MiniMax、DeepSeek,或任何兼容 OpenAI 的 API)。
工作原理
HermesGod 连接到 LLM API,并向其输入完整的世界状态:每个玩家的位置、生命值、物品栏、盔甲、生物群系、进度成就,甚至包括速度和光照等级。AI 以自然语言进行思考,然后输出 JSON 命令并在服务器上执行。
这意味着每次响应都是独一无二的——没有脚本,没有硬编码的恶作剧。AI 会即时决定做什么。
模式
? 生态模式 — 守护者
在聊天中与 HermesGod 对话。请求装备、资源、生物,或者只是闲聊。它可以:
- 给予你物品和装备
- 召唤生物供你训练
- 在你周围建造结构
- 解释游戏机制
- 或者只是聊天而不执行任何操作
"我需要铁甲" → 你获得铁甲
"教我 PvP" → 出现教程和练习用生物
"随便聊聊" → 它只说话,不执行命令
? 恶作剧模式 — 骗子
AI 会以动态难度静默地捉弄玩家,难度随玩家的进度而调整:
| 等级 |
发生的情况 |
| 1 |
令人毛骨悚然的声音、传送门粒子效果、灯光闪烁、令人迷失方向的传送 |
| 2 |
怪物伏击、物品被盗、负面效果、南瓜头 |
| 3 |
岩浆陷阱、黑曜石牢笼、唤魔者尖牙、TNT、精英生物 |
| 4 |
凋灵、监守者、虚空掉落、全面入侵 |
难度是根据游玩天数、盔甲、进度成就和维度来计算的。速通玩家会更快面临难度升级。新出生的玩家则会受到轻微的惊吓。
AI 会记住它上次做了什么(最多 10 次动作),以避免重复。
其他模式(开发中)
开发中...
命令
| 命令 |
作用 |
/hermesgod mode <eco/prank/disabled> |
切换模式 |
/hermesgod chat <message> |
从控制台与 HermesGod 对话 |
/hermesgod status |
查看当前模式、API 状态、最后一次动作 |
/hermesgod interval <seconds> |
设置固定的恶作剧思考间隔(例如 interval 60 = 每 60 秒一次)。设置为 0 以禁用 |
/hermesgod interval <min> <max> |
设置最小和最大秒数之间的随机间隔(例如 interval 30 90) |
/hermesgod api <key> |
在游戏中设置你的 API 密钥 |
/hermesgod url <endpoint> |
在游戏中设置 API 端点 URL |
/hermesgod model <name> |
设置 LLM 模型名称 |
/hermesgod maxtokens <count> |
设置每次 API 调用的最大令牌数(1~32000) |
/hermesgod reset |
清除对话历史 |
/hermesgod config |
查看所有当前设置 |
/hermesgod help |
显示所有命令 |
安装
- 安装 Minecraft Forge 1.20.1
- 将
.jar 文件放入 mods 文件夹
- 启动游戏 —— 会创建
config/hermesgod.json
- 使用你的 API 密钥和端点编辑配置文件:
{
"apiEndpoint": "https://api.openai.com/v1/chat/completions",
"apiKey": "sk-your-key-here",
"modelName": "gpt-4o-mini",
"currentMode": "eco"
}
- 重启游戏并输入
/hermesgod help
适用于单人游戏和专用服务器(仅服务端——客户端不需要安装模组)。
配置
所有设置均在 config/hermesgod.json 中:
| 设置项 |
默认值 |
描述 |
apiEndpoint |
https://api.minimaxi.com/v1/chat/completions |
兼容 OpenAI 的 API |
apiKey |
"" |
你的 API 密钥 |
modelName |
MiniMax-M2.7-highspeed |
LLM 模型 |
currentMode |
disabled |
起始模式 |
chatEnabled |
true |
允许聊天交互 |
prankIntervalMin |
30 |
恶作剧思考的最小间隔秒数(0 = 关闭) |
prankIntervalMax |
30 |
恶作剧思考的最大间隔秒数 |
maxTokens |
8000 |
每次 API 调用的最大令牌数 |
temperature |
0.8 |
AI 创造力 |
要求
- Minecraft 1.20.1 Forge (47.3.0+)
- Java 17+
- 你所选择的 LLM 提供商的 API 密钥
API 费用
HermesGod 会调用你选择的 LLM 提供商的 API。在生态模式中,仅当玩家发送聊天消息时调用。在恶作剧模式中,每 30 秒调用一次(可配置)。
使用轻量级模型如 gpt-4o-mini 或 MiniMax-M2.7-highspeed 以保持低成本。
常见问题
问:这是一个作弊客户端吗?
答:不是。它是一个使用 OP 命令的服务端模组。它不会给任何玩家带来不公平的优势。
问:我可以在多人服务器上用它吗?
答:可以。安装在服务器上即可。只需要一个 API 密钥。
问:为什么 AI 没有响应?
答:请检查你的 API 密钥、端点和网络连接。错误日志记录在 logs/latest.log 中。
问:我可以添加自己的恶作剧吗?
答:恶作剧系统是基于提示词驱动的。编辑 HermesBrain.java 以添加自定义的恶作剧描述。
HermesGod 注视。HermesGod 等待。HermesGod 决定。
HermesGod
An omniscient AI deity inhabits your Minecraft world — it sees everything players do, pranks them dynamically, helps with resources, or just chats.
What is HermesGod?
HermesGod is not a bot or an NPC. It's a world deity — a real LLM-powered intelligence that perceives your entire Minecraft world and reacts through server commands. Talk to it. Piss it off. Ask for help. It will respond.
It runs on any LLM you configure (OpenAI, Anthropic, MiniMax, DeepSeek, or any OpenAI-compatible API).
How It Works
HermesGod connects to an LLM API and feeds it the full world state: every player's position, health, inventory, armor, biome, advancements, even velocity and light level. The AI thinks in natural language, then outputs JSON commands that get executed on the server.
This means every response is unique — no scripts, no hardcoded pranks. The AI decides what to do on the fly.
Modes
? Eco Mode — Guardian
Talk to HermesGod in chat. Ask for gear, resources, mobs, or just chat. It can:
- Give you items and equipment
- Summon mobs for training
- Build structures around you
- Explain game mechanics
- Or just talk without doing anything
"I need iron armor" → You get iron armor
"Teach me PvP" → Tutorial + practice mobs appear
"Just chat" → It talks, no commands executed
? Prank Mode — Trickster
The AI silently pranks players with dynamic difficulty that scales with their progress:
| Level |
What Happens |
| 1 |
Creepy sounds, portal particles, flickering lights, disorienting teleports |
| 2 |
Monster ambushes, item theft, negative effects, pumpkin head |
| 3 |
Lava traps, obsidian prisons, evoker fangs, TNT, elite mobs |
| 4 |
Withers, wardens, void drops, full invasions |
Difficulty is calculated from days played, armor, advancements, and dimension. Speedrunners get escalated faster. Fresh spawns get gentle scares.
The AI remembers what it did last (up to 10 actions) to avoid repeating itself.
Other Modes (In Development)
In development...
Commands
| Command |
What it does |
/hermesgod mode <eco/prank/disabled> |
Switch mode |
/hermesgod chat <message> |
Talk to HermesGod from console |
/hermesgod status |
See current mode, API status, last action |
/hermesgod interval <seconds> |
Set fixed prank thinking interval (e.g. interval 60 = every 60s). Set 0 to disable |
/hermesgod interval <min> <max> |
Random interval between min and max seconds (e.g. interval 30 90) |
/hermesgod api <key> |
Set your API key in-game |
/hermesgod url <endpoint> |
Set API endpoint URL in-game |
/hermesgod model <name> |
Set the LLM model name |
/hermesgod maxtokens <count> |
Set max tokens per API call (1~32000) |
/hermesgod reset |
Clear conversation history |
/hermesgod config |
View all current settings |
/hermesgod help |
Show all commands |
Installation
- Install Minecraft Forge 1.20.1
- Put the
.jar in your mods folder
- Launch the game —
config/hermesgod.json is created
- Edit the config with your API key and endpoint:
{
"apiEndpoint": "https://api.openai.com/v1/chat/completions",
"apiKey": "sk-your-key-here",
"modelName": "gpt-4o-mini",
"currentMode": "eco"
}
- Restart and type
/hermesgod help
Works on single-player and dedicated servers (server-side only — clients don't need the mod).
Configuration
All settings in config/hermesgod.json:
| Setting |
Default |
Description |
apiEndpoint |
https://api.minimaxi.com/v1/chat/completions |
OpenAI-compatible API |
apiKey |
"" |
Your API key |
modelName |
MiniMax-M2.7-highspeed |
LLM model |
currentMode |
disabled |
Starting mode |
chatEnabled |
true |
Allow chat interaction |
prankIntervalMin |
30 |
Min seconds between prank thoughts (0 = off) |
prankIntervalMax |
30 |
Max seconds between prank thoughts |
maxTokens |
8000 |
Max tokens per API call |
temperature |
0.8 |
AI creativity |
Requirements
- Minecraft 1.20.1 Forge (47.3.0+)
- Java 17+
- An API key for your chosen LLM provider
API Costs
HermesGod calls your LLM provider's API. In Eco mode, only when a player sends a chat message. In Prank mode, every 30 seconds (configurable).
Use lightweight models like gpt-4o-mini or MiniMax-M2.7-highspeed to keep costs low.
FAQ
Q: Is this a hacked client?
A: No. It's a server-side mod that uses OP commands. It gives no unfair advantage to any player.
Q: Can I use it on my multiplayer server?
A: Yes. Install on the server. Only one API key needed.
Q: Why isn't the AI responding?
A: Check your API key, endpoint, and internet. Errors are logged in logs/latest.log.
Q: Can I add my own pranks?
A: The prank system is prompt-driven. Edit HermesBrain.java to add custom trick descriptions.
HermesGod watches. HermesGod waits. HermesGod decides.
评论 0
登录后参与评分
还没有评论,来抢沙发吧