Solver
你的 AI 驱动服务器助手
Solver 是一款基于 AI 的 Minecraft 服务器管理与审核插件,适用于 Paper/Folia,其核心是内置的 AI 助手 Fyrx。它最初只是一个崩溃报告翻译器——从此告别在数百行晦涩难懂的 Java 堆栈跟踪中手动查找问题——现已发展为一个完整的审核/员工替代方案:具备上下文感知的 AI 聊天审核、包含小号检测与游戏内申诉功能的完整制裁系统、员工报告队列,以及深度的 Staff Mode Toolkit(包括隐身、冻结、物品栏检查等)。所有功能均支持你选择的 AI 提供商:Anthropic Claude、Google Gemini 或任何兼容 OpenAI 的端点(包括本地/自托管模型)。
每当你的服务器发生错误、崩溃、卡顿尖峰或聊天事件时,Fyrx 会在后台进行分析,并在控制台或游戏中为你提供人类可读的解释和可操作的后续步骤。
主要功能
诊断工具
- AI 崩溃分析: 自动拦截原生 JVM 崩溃和标准 Minecraft 崩溃报告,明确指出是哪个插件、内存问题还是损坏的区块导致了灾难。
- 早期启动错误检测: 捕获服务器甚至尚未完全启动时就发生的插件不兼容错误,在启动时立即向你发出警报。
- 卡顿与冻结检测 (TickMonitor): 持续监控服务器的 TPS。如果主线程冻结超过几秒钟(死锁),Fyrx 会分析线程转储并告诉你哪个插件拖慢了服务器。原生支持 Folia,后者自带自己的看门狗机制。
AI 聊天审核
- 阅读整个对话,而非简单的词库黑名单: Fyrx 评估聊天交换的真实上下文,以区分持续的骚扰和朋友间无害的玩笑。
- 自定义分类: 裁决结果会根据你在
tags.yml 中配置的任何类别进行标记(默认附带 TOXIC、HARASSMENT、THREAT、HATE_SPEECH、SCAM、SPAM)——每个实际违规的玩家对应一个条目,而不是对整个扫描猜测单个单词。
- 置信度计量 + 多信号佐证: 每个裁决都会显示 Fyrx 的确定程度(0-100%),这与严重程度是分开的——真正的自动制裁需要高置信度 且 高严重程度,或者独立的佐证信号,因此单一的 AI 不确定调用不能单独导致某人受罚。
- 可配置的响应: 仅警报、私人警告,或根据严重程度或类别执行全自动静音/踢出/封禁 —— 使用
tag-overrides 将任何标签直接映射到制裁类型,这也是实现聊天审核自动封禁的唯一途径。
- 已审查过的文本不会从头重新评判 —— 一旦消息被清除,它仅作为背景上下文用于识别升级模式,绝不会无缘无故地被重新标记。
制裁系统 (BETA)
/solver warn|mute|tempmute|kick|ban|tempban <player> [duration] <reason> —— 每项制裁都拥有唯一的 ID 和持久化的历史记录(/solver history、/solver check <id>、/solver sanctions <player> GUI)。
- 封禁在重新连接时真正生效,而不仅仅是一次性的踢出。即使关闭了 AI 聊天审核,静音依然有效。
- 真实的小号/IP 交叉引用:
/solver checkuser <player> 显示曾与该玩家共享 IP 的所有账号。封禁会立即踢出已经在线的小号,而不仅仅是阻止未来的登录。
- 游戏内制裁申诉:
/solver appeal <id> <reason> —— 无需 Discord webhook。员工会立即收到通知并使用 /solver appeal list|accept|reject 进行审核。
- 原因模板与自动警告升级: 使用
#name 快速输入常见原因,并在玩家受到第 N 次警告后自动踢出/临时封禁/执行命令。
- 默认 SQLite,需要时使用 MySQL: 零配置的本地存储,或将
punishments.storage 指向共享的 MySQL 数据库以支持多服务器网络。
员工报告系统
/solver report <player> <reason> 允许任何玩家标记内容以供员工关注 —— 对所有人开放,完全独立于制裁系统。
/solver reports [claim|close|reopen|gui] 管理队列及其专属 GUI;当报告被认领或关闭时,举报人会直接收到通知。
Staff Mode Toolkit (BETA)
- 隐身、冻结和员工聊天: 两级隐身(
strict 级别连大多数员工也看不见),隐身期间可选的伤害免疫,以及私有的冻结聊天频道,以便在屏幕共享期间真正与某人交谈,而不仅仅是使其沉默。
- CommandSpy: 其他玩家的命令实时中继到你的聊天窗口,而不仅仅是事后记录。
- 命名组合配置文件:
/solver staffmode <profile> 将隐身 + 冻结免疫 + 飞行 + 上帝模式 + 清空物品栏(自动保存/恢复)打包成一个切换开关。
/solver fly/god/rtp: 独立的飞行和无敌能力(真正的 PvP -proof 无敌,而不仅仅是一个标志),以及随机传送至非员工玩家处进行监督巡逻。
/solver inspect/enderchest: 查看并没收玩家的物品栏或末影箱,而无需打开它们。
- 上述所有命令也可以不带
/solver 前缀使用(如 /vanish、/freeze、/report 等)——如果其他插件已经占用了这些名称,你可以在 config.yml 中关闭任何类别。
完全可自定义的消息
- 插件发送的所有消息(帮助、错误、制裁、管理警报)均保存在
messages.yml 中,并完整支持 MiniMessage —— 颜色、渐变、悬停文本,随心所欲。
- 开箱即用的多语言支持,涵盖 9 种区域设置:
en_US, es_ES, pt_BR, ru_RU, de_DE, fr_FR, ja_JP, ko_KR, eo_EO。
可选集成 (BETA)
-
PlaceholderAPI 支持 — %solver_muted%、%solver_warns%、%solver_active_sanctions% 等,适用于你的记分板/标签页/聊天插件。如果你未安装 PlaceholderAPI,则此功能无效。
-
简洁美观的 UI: AI 响应和所有游戏内菜单均采用原生 Minecraft 颜色代码 / MiniMessage 格式,直接在控制台和游戏内显示。
-
100% 异步 & 零延迟: 所有 AI 处理、文件读取、数据库访问和网络请求均在主线程之外运行。当 Fyrx 思考时,你的服务器 TPS 不会下降。

使用方法
- 将
Solver-0.8.0.jar 放入你的 plugins 文件夹。
- 启动服务器一次以生成
config.yml。
- 打开
plugins/Solver/config.yml,选择提供商(anthropic、google 或 other),并粘贴其 API 密钥。
- 重启服务器。Fyrx 现在正在监控!在游戏中或控制台中运行
/solver help 查看完整的命令列表。
(想安全地测试它?使用 /solver crashme dry-run 进行合成崩溃分析,或使用 /solver moderation test <message> 尝试管理 AI —— 这两者都不会影响你的真实服务器。)
要求
- 服务器类型: Paper, Folia, Purpur 或其分支。(聊天管理需要 Paper 的
AsyncChatEvent —— Spigot 尚不支持,已在路线图计划中)。
- Minecraft 版本: 1.18.x — 1.21.x。
- Java: Java 17 或更高版本。
- API 密钥: 来自你选择的提供商 —— Anthropic、Google Gemini(有免费套餐)或任何 OpenAI 兼容端点。
- 可选: EssentialsX 会被自动检测 —— 如果已安装 EssentialsX,
/enderchest 命令不会单独注册,因此这两个插件永远不会争夺相同的命令名称。

config.yml
# ███████╗ ██████╗ ██╗ ██╗ ██╗███████╗██████╗
# ██╔════╝██╔═══██╗██║ ██║ ██║██╔════╝██╔══██╗
# ███████╗██║ ██║██║ ██║ ██║█████╗ ██████╔╝
# ╚════██║██║ ██║██║ ██║ ██║██╔══╝ ██╔══██╗
# ███████║╚██████╔╝███████╗╚██████╔╝███████╗██║ ██║
# ╚══════╝ ╚═════╝ ╚══════╝ ╚═════╝ ╚══════╝╚═╝ ╚═╝
# Solver Configuration
# Language (en_US, es_ES, pt_BR, ru_RU, de_DE, fr_FR, ja_JP, ko_KR, eo_EO)
localization: en_US
# Automatically analyze whether there was a crash when the server starts up
check-crash-on-startup: true
# Monitor the entire console for errors (red text)
monitor-console: true
# Avoid analyzing the same error multiple times (minutes to wait)
error-cooldown-minutes: 10
# Solver Main Configuration
# Show the giant ASCII art banner in the console on startup?
show-banner: true
# AI Provider Configuration
ai-provider:
# Provider: 'anthropic', 'google' or 'other'
provider: "anthropic"
# Model to use
# Anthropic: Sonnet, Haiku, Opus
# Google: Pro, Flash
# Other: Model name (e.g: gpt-4o, mistral-large)
model: "Sonnet"
# API Keys
anthropic-key: ""
google-key: ""
# Configuration for the 'other' provider (OpenAI-compatible)
other-key: ""
other-url: "https://api.openai.com/v1/chat/completions"
# Analysis configuration
analysis:
# Automatically analyze exceptions
auto-analyze: true
# Timeout for analysis (seconds)
timeout: 60
# Save analysis to file
save-to-file: true
# Reports directory
reports-directory: "crash-reports"
# How many days to keep analyses saved in analysis-logs/ before deleting them.
# 0 = keep forever.
reports-retention-days: 0
# Notifications
notifications:
# Notify admins in-game when there's an error
notify-admins: true
# Send analysis to chat (in addition to console)
send-to-chat: false
# AI Chat Moderation (Fyrx reads chat and evaluates the CONTEXT of the conversation,
# not a word blacklist. Enabling this sends chat content to the AI provider configured
# above in 'ai-provider'. Disabled by default for privacy.
chat-moderation:
enabled: false
# If true (default), OPs are automatically exempt from chat moderation.
# Set to false so OPs are also examined like any other player. The
# solver.moderation.bypass permission still works independently to exempt a specific
# player (OP or not), regardless of this value.
exempt-ops: true
# Maximum recent messages to remember per player / globally
buffer-size: 50
# Time window of context sent to the AI (minutes)
context-window-minutes: 5
# How often (seconds) to check whether there's new chat to analyze (unconditional
# sweep, never depends on the pre-filter: nothing can be hidden from the AI this way)
analysis-interval-seconds: 45
# How to scan chat each cycle:
# bulk: 1 single AI call with the whole mixed conversation. Cheap, but with many
# players talking at once the AI can confuse who said what.
# individual: 1 AI call PER player with new activity. The offender is already known
# beforehand (the AI doesn't have to guess), so this eliminates that confusion
# risk, but the cost scales with how many players are active per cycle - on a
# busy server this can burn through the provider's rate limit fast.
scan-mode: "bulk"
AI 实际被调用来检查聊天的频率 (0-3)。上面的 scan-mode 决定了 HOW(如何)调用 AI;这里决定 WHEN(何时)调用。
0 = 从不调用 AI - 上面的预过滤器完全独立决策(更便宜/更快,
准确性较低;即使未配置任何 AI 提供商也能工作)。
1 = 仅当预过滤器已标记某些可疑内容时才调用 AI(平衡方案)。
2 = 默认/经典行为:AI 每个周期都会审查所有新聊天,无论
预过滤器是否标记了任何内容。
3 = 实时:每条消息在到达时立即发送给 AI,而不是
等待 analysis-interval-seconds(反应速度更快,但代价是调用次数大幅增加)。
llm-analysis-level: 2
轻量级预过滤器:计算紧迫度分数 (0.0-1.0),结合多个信号。
永远不会决定某事是否是骚扰或阻止分析 - 上面的扫描
仍然会读取所有聊天。它仅决定是否值得提前唤醒
Fyrx 而不是等待完整间隔,以便在明显情况下更快做出反应。
pre-filter:
# 如果同一玩家连续收到来自他人的多条消息,则提高紧迫度分数
# (可能对固定目标进行持续的骚扰)。
trigger-on-repeated-target: true
# 大写消息的比例 (0.0-1.0),超过该比例被视为“大喊”
# 并增加紧迫度分数 (0.6 = 60% 或更多的消息为大写)。
trigger-on-caps-ratio: 0.6
# 需要积累多少条来自同一玩家的消息,预过滤器才开始
# 评估上述的“重复目标”模式。
min-messages-before-trigger: 3
# 触发即时分析而不是等待的最小分数 (0.0-1.0)
urgent-score-threshold: 0.65
当 AI 报告有毒/骚扰判定时的操作
action:
# alert-only:仅通知拥有 solver.notify 权限的玩家
# warn-player:alert-only + 向被标记的玩家发送私人通知
# auto-action:warn-player + 基于严重程度 (1-5) 自动静音/踢出
mode: "alert-only"
# “auto-action” 自动静音玩家而不是仅仅警告他们的严重程度 (1-5,由 AI 在每个判定中设置)。
severity-threshold-mute: 4
# “auto-action” 自动踢出玩家而不是静音他们的严重程度。
# 应该 >= severity-threshold-mute 以使升级合理。
severity-threshold-kick: 5
# “auto-action” 在超过
# severity-threshold-mute 时应用的自动静音持续时间(分钟),不影响手动 /solver mute|tempmute。
mute-duration-minutes: 10
# 可选:为特定的“情况”(AI 标签)强制特定的制裁类型,无论
# 上面的 severity-threshold-mute/kick 如何 - 仅在 mode 为 "auto-action" 时使用。有效标签:
# TOXIC, HARASSMENT, THREAT, HATE_SPEECH, SCAM, SPAM。有效类型:warn, mute, tempmute, kick,
# ban, tempban(tempmute/tempban 需要 'duration',格式与其他地方相同 s/m/h/d/w)。
# 如果一个判定匹配多个配置的标签,最严厉的配置类型获胜。这也
# 是通过聊天审核实现自动封禁/临时封禁的唯一方式 - 严重程度
# 阈值本身永远不会超过踢出。
tag-overrides: {}
# THREAT:
# type: tempban
# duration: 7d
# HATE_SPEECH:
# type: tempban
# duration: 3d
# SCAM:
# type: ban
# 多信号佐证:仅当 AI 判定的置信度至少为此值且严重程度至少为此值时,才能触发真正的警告/静音/踢出
# 低于此值时,还需要至少一个佐证的预过滤器信号(大写/侮辱/重复目标/爆发 - 见上面的 pre-filter);
# 如果没有这样的信号,它仅作为员工警报供通过 /solver check 进行手动审核,绝不会成为
# 自动制裁。置信度 0-100,严重程度 1-5。
min-confidence-to-act-alone: 85
min-severity-to-act-alone: 4
将每次检测到的事件保存到 moderation-reports/
log-incidents-to-file: true
当玩家运行 /solver moderation test <message> 时,也在控制台显示结果
(如果是控制台运行命令,控制台始终已经看到它)。
log-test-results-to-console: true
在删除之前保留 moderation-reports/ 中保存的事件的天数。
0 = 永久保留。
reports-retention-days: 0
这些命令组是否也可以在不使用 /solver 前缀的情况下直接运行(例如 /vanish
而不是 /solver vanish)。无论如何,/solver <command> 始终有效。如果你安装的其他插件已经占用了一个
其命令名称(例如 EssentialsX 也有 /mute, /kick, /ban, /warn, /tempban, /tempmute,
/unmute, /unban)- 否则 Bukkit 会让先注册的那个插件保留裸名。
/enderchest 是唯一自动处理的例外:如果检测到 EssentialsX,其
独立形式会自动跳过(无需为此关闭整个类别)- EssentialsX 的 /enderchest 意思是“显示 MY OWN enderchest”,含义不同。
standalone-commands:
staff-mode: true # /vanish, /freeze, /staffchat, /commandspy, /enderchest(除非安装了 EssentialsX)
sanctions: true # /warn, /mute, /tempmute, /kick, /ban, /tempban, /unban, /unmute, /unwarn, /history, /check, /note, /checkuser, /appeal, /sanctions
moderation: true # /moderation
reports: true # /report
将已执行的命令实时转发给启用了 /solver commandspy 的用户——不仅仅是
事后记录。
commandspy:
enabled: true
"*" 监听所有命令。否则列出特定命令名称(不带前导的 /)以仅
监听那些命令,例如 ["op", "gamemode", "give"]。
watch-patterns: ["*"]
无论谁在监听,都不转发的玩家名称(不区分大小写)。
exempt-players: []
持有以下任一权限的玩家也永远不会被转发。
exempt-permissions: []
⚠ BETA:制裁系统(通过 /solver 进行警告/静音/踢出/封禁,sanctions.db 中持久化历史记录)。
在你完全熟悉它在服务器上的行为之前,请在每次制裁后仔细检查 /solver history。
punishments:
存储制裁历史的位置:sqlite(默认,零配置,本地文件)或 mysql
(用于共享同一个制裁数据库的多服务器设置)。
storage: sqlite
仅在 storage 为 mysql 时读取。
mysql:
host: localhost
port: 3306
database: solver
user: solver
password: ""
快捷原因模板:"/solver ban Player #griefing" 会展开为配置好的文本
(如果制裁类型需要持续时间且未明确提供,则包括持续时间)。未匹配的
#名称将作为字面原因保留,因此这绝不会破坏真正以 "#" 开头的原因。
reason-templates: {}
griefing:
text: "Griefing / destroying other players' builds"
duration: "1d"
警告计数自动升级:在记录 WARN 后,如果玩家的总警告计数与此处的键匹配,
配置的动作用于自动触发。"action: sanction" 应用另一个制裁(类型/原因/持续时间,与手动相同);
"action: command" 运行控制台命令({player} 替换为玩家名称)。
warn-escalation: {}
"3":
action: sanction
type: kick
reason: "Auto: 3 warnings"
"5":
action: sanction
type: tempban
duration: "1d"
reason: "Auto: 5 warnings"
appeals:
# 在 warn/mute/kick/ban 消息中显示指向 /solver appeal <id> <reason> 的行。
show-in-sanction-message: true
⚠ BETA:员工模式工具包(通过 /solver 实现隐身/冻结/员工聊天)。
staff-mode:
vanish:
# 从服务器列表(MOTD)中显示的在线玩家计数中减去隐身的玩家
hide-from-server-list: true
# 使隐身玩家在隐身期间免受伤害。
invulnerable-while-vanished: false
freeze:
# 允许在玩家被冻结时使用的命令(不带前导的 /)
allowed-commands: ["msg", "tell", "r", "helpop"]
为 /solver staffmode <profile> 命名的组合配置文件——每个配置文件同时切换一组
独立的员工工具,而不是逐个打开它们。每个字段都是可选的,默认为关闭。clear-inventory 始终先保存你的物品,并在你离开配置文件时恢复它们。
profiles: {}
moderator:
vanish: true
freeze-immunity: true
flight: true
god-mode: true
clear-inventory: true
builder:
flight: true
/solver inspect <player> - 在不实际打开的情况下查看玩家的物品栏。
inspect:
每隔几刻刷新一次打开的物品栏,而不是一次性快照。这是轮询,不是真正的推送更新(Bukkit 没有“物品栏更改”事件可供响应)。
live-refresh: false
live-refresh-interval-ticks: 10
</details>
<details>
<summary>Commands</summary>
Base command: `/solver` (alias: `/as`). Most subcommands also work without the `/solver` prefix — see `standalone-commands` in `config.yml`.
## Commands
/solver help:显示可用命令列表。
/solver reload:重新加载插件的配置和消息。
/solver analyze-last:重新分析上次检测到的崩溃/错误。
/solver crashme <exception|deadlock|oom|dry-run>:触发一个合成错误,用于测试 AI 分析管道。
/solver moderation status:显示当前的聊天审核诊断信息(缓冲区、最近裁决、行动模式、扫描模式、AI 分析级别等)。
/solver moderation test <message>:将一条消息通过 AI 审核提示进行处理,而不作为实际的聊天消息发送。
/solver warn <player> <reason>:发出警告。原因参数接受来自 reason-templates 的 #模板。
/solver mute <player> <reason>:永久静音一名玩家。
/solver tempmute <player> <duration> <reason>:在指定时长内静音一名玩家。
/solver kick <player> <reason>:踢出一名玩家。
/solver ban <player> <reason>:永久封禁一名玩家。同时会立即踢出任何已知已在线的小号。
/solver tempban <player> <duration> <reason>:在指定时长内封禁一名玩家。
/solver unban <player>:解除当前有效的封禁/临时封禁。
/solver unmute <player>:解除当前有效的静音/临时静音。
/solver unwarn <player>:撤销该玩家最近一次有效的警告。
/solver history <player>:显示某玩家的完整处罚历史。
/solver check <id>:根据 ID 显示单个处罚的详细信息。
/solver note <player> <text>:为某玩家附加一条仅工作人员可见的备注,不发出处罚。
/solver checkuser <player>:交叉引用 IP 历史记录——显示该网络下所有已知的小号账号以及任何正在生效的处罚。
/solver appeal <id> <reason>:允许受罚玩家在游戏中申诉自己的处罚。默认情况下对所有人开放。
/solver appeal list|accept|reject <id>:工作人员审核待处理的申诉。
/solver sanctions <player>:打开 GUI 浏览某玩家的完整处罚历史;点击其中一项有效处罚可获取具体的撤销命令。
/solver report <player> <reason>:提交一份针对某玩家的报告以供工作人员关注。默认情况下对所有人开放。
/solver reports:列出未处理的报告。
/solver reports claim|close|reopen <id> [reason]:管理报告队列。通知原始举报者。
/solver reports gui:打开未处理报告的 GUI(左键点击认领,右键点击关闭)。
/solver vanish [strict]:切换自身的隐身状态(对其他玩家、怪物锁定、加入/离开消息、Tab 补全以及服务器列表计数隐藏)。“strict” 是第二层隐身状态,大多数工作人员也无法看到。
/solver freeze <player>:切换对某玩家的冻结状态(无法移动、无法运行大多数命令、无法破坏/放置方块、无法造成或承受战斗伤害)。即使是被冻结的玩家自己也可以随时切换此状态。
/solver freeze <player> <message>:在该玩家私有的冻结聊天频道中发送一条消息,而不是切换冻结状态。
/solver staffchat:切换工作人员聊天模式(你的普通聊天消息将仅发送给工作人员,而非公共聊天)。
/solver staffchat <message>:向工作人员聊天发送一条一次性消息,而不切换模式。
/solver commandspy:切换实时中继其他玩家执行的命令到你的聊天窗口。
/solver staffmode <profile>:切换一个命名的组合配置文件(隐身/冻结免疫/飞行/上帝模式/清空物品栏),该配置在 config.yml 中定义。
/solver fly:切换飞行状态,独立于任何配置文件。
/solver god:切换无敌状态,包括 PvP 伤害。
/solver rtp:将你传送至一名随机的在线非工作人员玩家处。
/solver inspect <player>:查看某玩家的物品栏/护甲/副手物品,而无需实际打开界面;点击物品即可没收它。
/solver enderchest <player>:对某玩家的末影箱执行相同的查看/没收操作。
</details>
<details>
<summary>权限</summary>
## 权限
solver.admin:完全访问以下所有命令。默认:op。
solver.notify:接收错误和聊天审核的工作人员警报。默认:op。
solver.moderation.bypass:无论 OP 状态如何,豁免特定玩家接受聊天审核。默认:false。
solver.diagnostics.*:所有诊断命令(reload/analyze-last/crashme/moderation)。默认:false。
solver.diagnostics.reload:允许使用 /solver reload。
solver.diagnostics.analyzelast:允许使用 /solver analyze-last。
solver.diagnostics.crashme:允许使用 /solver crashme。
solver.diagnostics.moderation:允许使用 /solver moderation status|test。
solver.sanctions.*:所有处罚命令。默认:false。
solver.sanctions.warn:允许使用 /solver warn。
solver.sanctions.mute:允许使用 /solver mute。
solver.sanctions.tempmute:允许使用 /solver tempmute。
solver.sanctions.kick:允许使用 /solver kick。
solver.sanctions.ban:允许使用 /solver ban。
solver.sanctions.tempban:允许使用 /solver tempban。
solver.sanctions.unban:允许使用 /solver unban。
solver.sanctions.unmute:允许使用 /solver unmute。
solver.sanctions.unwarn:允许使用 /solver unwarn。
solver.sanctions.history:允许使用 /solver history。
solver.sanctions.check:允许使用 /solver check。
solver.sanctions.note:允许使用 /solver note。
solver.sanctions.checkuser:允许使用 /solver checkuser。
solver.sanctions.appeal:允许使用 /solver appeal <id> <reason>(提交自己的申诉)。默认:true。
solver.sanctions.appeal.manage:允许使用 /solver appeal list|accept|reject。
solver.sanctions.gui:允许使用 /solver sanctions(历史 GUI)。
solver.reports.*:所有工作人员报告命令。默认:false。
solver.report:允许使用 /solver report(提交报告)。默认:true。
solver.reports.manage:允许使用 /solver reports 列表/认领/关闭/重新打开。
solver.reports.gui:允许使用报告 GUI。
solver.staffmode.*:所有 Staff Mode Toolkit 命令。默认值:false。
solver.staffmode.vanish:允许使用 /solver vanish。
solver.staffmode.vanish.see-strict:可看穿其他玩家的严格层级隐身状态。
solver.staffmode.freeze:允许使用 /solver freeze。
solver.staffmode.staffchat:允许使用 /solver staffchat。
solver.staffmode.commandspy:允许使用 /solver commandspy。
solver.staffmode.profiles:允许使用 /solver staffmode <profile>。
solver.staffmode.fly:允许使用 /solver fly。
solver.staffmode.god:允许使用 /solver god。
solver.staffmode.rtp:允许使用 /solver rtp。
solver.staffmode.silentjoin:加入时自动隐身且不显示加入消息。
solver.staffmode.inspect:允许使用 /solver inspect。
solver.staffmode.enderchest:允许使用 /solver enderchest。
</details>
---
*由 FyrxLab 用心制作 ♥️*
Solver
Your AI-Powered Server Assistant
Solver is the AI-powered Minecraft server admin and moderation plugin for Paper/Folia, built around Fyrx, your built-in AI assistant. It started as a crash-report translator — say goodbye to scrolling through hundreds of lines of cryptic Java stack traces — and has grown into a full moderation/staff alternative: contextual AI chat moderation, a complete sanctions system with alt-account detection and in-game appeals, a staff reports queue, and a deep Staff Mode Toolkit (vanish, freeze, inventory inspection, and more). Everything runs on the AI provider of your choice: Anthropic Claude, Google Gemini, or any OpenAI-compatible endpoint (including local/self-hosted models).
Whenever your server hits an error, a crash, a lag spike, or a chat incident, Fyrx analyzes it in the background and gives you a human-readable explanation and actionable next steps directly in your console or in-game.
Key Features
Diagnostics
- AI Crash Analysis: Automatically intercepts native JVM crashes and standard Minecraft crash reports, explaining exactly which plugin, memory issue, or broken chunk caused the disaster.
- Early Startup Error Detection: Captures plugin incompatibility errors that happen before the server even fully starts, alerting you immediately upon boot.
- Lag & Freeze Detection (TickMonitor): Continuously monitors your server's TPS. If the main thread freezes for more than a few seconds (deadlocks), Fyrx analyzes the thread dump and tells you which plugin is choking the server. Runs natively on Folia, which handles its own watchdog.
AI Chat Moderation
- Reads the whole conversation, not a word blacklist: Fyrx evaluates the real CONTEXT of a chat exchange to tell insistent harassment apart from harmless banter between friends.
- Categories you define: verdicts come back tagged with whatever categories you configure in
tags.yml (ships with TOXIC, HARASSMENT, THREAT, HATE_SPEECH, SCAM, SPAM) — one entry per player actually misbehaving, not a single guessed word for the whole scan.
- Confidence meter + multi-signal corroboration: every verdict shows how sure Fyrx is (0-100%), separate from severity — and a real automatic sanction requires either high confidence and severity, or an independent corroborating signal, so a single shaky AI call can't get someone punished alone.
- Configurable response: alert-only, private warning, or full auto mute/kick/ban by severity or by category — map any tag straight to a sanction type with
tag-overrides, including the only path to an automatic ban from chat moderation.
- Already-reviewed text never gets re-judged from scratch — once a message is cleared, it's only used as background context for spotting an escalating pattern, never re-flagged out of nowhere.
Sanctions System (BETA)
/solver warn|mute|tempmute|kick|ban|tempban <player> [duration] <reason> — every sanction gets a real ID and a persistent history (/solver history, /solver check <id>, /solver sanctions <player> GUI).
- Bans are actually enforced on reconnect, not just a one-time kick. Mutes work even with AI chat moderation turned off.
- Real alt-account/IP cross-referencing:
/solver checkuser <player> shows every account that ever shared an IP with it. A ban instantly kicks an alt that's already online, not just future logins.
- In-game sanction appeals:
/solver appeal <id> <reason> — no Discord webhook needed. Staff get notified immediately and review with /solver appeal list|accept|reject.
- Reason templates & automatic warn escalation: shortcut a common reason with
#name, and auto-kick/tempban/run a command after a player's Nth warning.
- SQLite by default, MySQL when you need it: zero-config local storage, or point
punishments.storage at a shared MySQL database for multi-server networks.
Staff Reports System
/solver report <player> <reason> lets any player flag something for staff attention — open to everyone, entirely separate from sanctions.
/solver reports [claim|close|reopen|gui] manages the queue with its own GUI; the reporter is notified directly when their report is claimed or closed.
Staff Mode Toolkit (BETA)
- Vanish, freeze, and staff chat: two-tier vanish (a
strict level invisible even to most staff), optional damage immunity while vanished, and a private freeze chat channel to actually talk to someone during a screenshare instead of just silencing them.
- CommandSpy: a live relay of other players' commands into your chat, not just after-the-fact logging.
- Named combined profiles:
/solver staffmode <profile> bundles vanish + freeze-immunity + flight + god-mode + inventory-clear (auto saved/restored) into one toggle.
/solver fly/god/rtp: independent flight and invulnerability (real PvP-proof invulnerability, not just a flag), plus a random teleport to a non-staff player for supervision rounds.
/solver inspect/enderchest: view and confiscate a player's inventory or ender chest without ever opening it.
- Every command above also works without the
/solver prefix (/vanish, /freeze, /report, etc.) — toggle any category off in config.yml if another plugin already claims that name.
Fully Customizable Messages
- Every message the plugin sends (help, errors, sanctions, moderation alerts) lives in
messages.yml with full MiniMessage support — colors, gradients, hover text, whatever you want.
- Multi-language out of the box across 9 locales:
en_US, es_ES, pt_BR, ru_RU, de_DE, fr_FR, ja_JP, ko_KR, eo_EO.
Optional Integrations (BETA)
-
PlaceholderAPI support — %solver_muted%, %solver_warns%, %solver_active_sanctions%, and more for your scoreboard/tab-list/chat plugins. Does nothing if you don't have PlaceholderAPI installed.
-
Clean & Beautiful UI: AI responses and every in-game menu are formatted using native Minecraft color codes / MiniMessage directly in your console and in-game.
-
100% Asynchronous & Zero-Lag: All AI processing, file reading, database access, and network requests run off the main thread. Your server's TPS won't drop while Fyrx is thinking.

How to use
- Drop
Solver-0.8.0.jar into your plugins folder.
- Start the server once to generate
config.yml.
- Open
plugins/Solver/config.yml, pick a provider (anthropic, google, or other), and paste its API key.
- Restart the server. Fyrx is now monitoring! Run
/solver help in-game or console for the full command list.
(Want to test it safely? Use /solver crashme dry-run for a synthetic crash analysis, or /solver moderation test <message> to try the moderation AI — neither one touches your real server.)
Requirements
- Server Type: Paper, Folia, Purpur, or forks. (Chat moderation needs Paper's
AsyncChatEvent — Spigot isn't supported yet, it's on the roadmap).
- Minecraft Version: 1.18.x — 1.21.x.
- Java: Java 17 or higher.
- API Key: From your chosen provider — Anthropic, Google Gemini (has a free tier), or any OpenAI-compatible endpoint.
- Optional: EssentialsX is auto-detected —
/enderchest won't register standalone if it's installed, so the two plugins never fight over the same command name.

config.yml
# ███████╗ ██████╗ ██╗ ██╗ ██╗███████╗██████╗
# ██╔════╝██╔═══██╗██║ ██║ ██║██╔════╝██╔══██╗
# ███████╗██║ ██║██║ ██║ ██║█████╗ ██████╔╝
# ╚════██║██║ ██║██║ ██║ ██║██╔══╝ ██╔══██╗
# ███████║╚██████╔╝███████╗╚██████╔╝███████╗██║ ██║
# ╚══════╝ ╚═════╝ ╚══════╝ ╚═════╝ ╚══════╝╚═╝ ╚═╝
# Solver Configuration
# Language (en_US, es_ES, pt_BR, ru_RU, de_DE, fr_FR, ja_JP, ko_KR, eo_EO)
localization: en_US
# Automatically analyze whether there was a crash when the server starts up
check-crash-on-startup: true
# Monitor the entire console for errors (red text)
monitor-console: true
# Avoid analyzing the same error multiple times (minutes to wait)
error-cooldown-minutes: 10
# Solver Main Configuration
# Show the giant ASCII art banner in the console on startup?
show-banner: true
# AI Provider Configuration
ai-provider:
# Provider: 'anthropic', 'google' or 'other'
provider: "anthropic"
# Model to use
# Anthropic: Sonnet, Haiku, Opus
# Google: Pro, Flash
# Other: Model name (e.g: gpt-4o, mistral-large)
model: "Sonnet"
# API Keys
anthropic-key: ""
google-key: ""
# Configuration for the 'other' provider (OpenAI-compatible)
other-key: ""
other-url: "https://api.openai.com/v1/chat/completions"
# Analysis configuration
analysis:
# Automatically analyze exceptions
auto-analyze: true
# Timeout for analysis (seconds)
timeout: 60
# Save analysis to file
save-to-file: true
# Reports directory
reports-directory: "crash-reports"
# How many days to keep analyses saved in analysis-logs/ before deleting them.
# 0 = keep forever.
reports-retention-days: 0
# Notifications
notifications:
# Notify admins in-game when there's an error
notify-admins: true
# Send analysis to chat (in addition to console)
send-to-chat: false
# AI Chat Moderation (Fyrx reads chat and evaluates the CONTEXT of the conversation,
# not a word blacklist. Enabling this sends chat content to the AI provider configured
# above in 'ai-provider'. Disabled by default for privacy.
chat-moderation:
enabled: false
# If true (default), OPs are automatically exempt from chat moderation.
# Set to false so OPs are also examined like any other player. The
# solver.moderation.bypass permission still works independently to exempt a specific
# player (OP or not), regardless of this value.
exempt-ops: true
# Maximum recent messages to remember per player / globally
buffer-size: 50
# Time window of context sent to the AI (minutes)
context-window-minutes: 5
# How often (seconds) to check whether there's new chat to analyze (unconditional
# sweep, never depends on the pre-filter: nothing can be hidden from the AI this way)
analysis-interval-seconds: 45
# How to scan chat each cycle:
# bulk: 1 single AI call with the whole mixed conversation. Cheap, but with many
# players talking at once the AI can confuse who said what.
# individual: 1 AI call PER player with new activity. The offender is already known
# beforehand (the AI doesn't have to guess), so this eliminates that confusion
# risk, but the cost scales with how many players are active per cycle - on a
# busy server this can burn through the provider's rate limit fast.
scan-mode: "bulk"
# How often the AI actually gets called to check chat (0-3). scan-mode above decides HOW
# the AI is called; this decides WHEN.
# 0 = never call the AI - the pre-filter above decides entirely on its own (cheaper/faster,
# less accurate; works with no AI provider configured at all).
# 1 = only call the AI once the pre-filter already flagged something suspicious (balanced).
# 2 = default/classic behavior: the AI reviews all new chat every cycle, regardless of
# whether the pre-filter flagged anything.
# 3 = real-time: every message is sent to the AI immediately as it arrives, instead of
# waiting for analysis-interval-seconds (much faster reaction, at the cost of many more calls).
llm-analysis-level: 2
# Lightweight pre-filter: computes an urgency score (0.0-1.0) combining several signals.
# NEVER decides whether something is harassment nor blocks the analysis - the sweep
# above still reads all the chat regardless. It only decides whether it's worth WAKING
# Fyrx up early instead of waiting for the full interval, to react faster in obvious cases.
pre-filter:
# Raises the urgency score if the same player receives several messages in a row from
# others (possible insistent harassment toward a fixed target).
trigger-on-repeated-target: true
# Fraction (0.0-1.0) of a message in UPPERCASE from which it's considered "shouted"
# and adds urgency score (0.6 = 60% or more of the message in caps).
trigger-on-caps-ratio: 0.6
# How many messages from a player need to build up before the pre-filter starts
# evaluating the "repeated target" pattern described above.
min-messages-before-trigger: 3
# Minimum score (0.0-1.0) to trigger an immediate analysis instead of waiting
urgent-score-threshold: 0.65
# What to do when the AI reports a toxic/harassment verdict
action:
# alert-only: only notifies players with the solver.notify permission
# warn-player: alert-only + a private notice to the flagged player
# auto-action: warn-player + automatic mute/kick based on severity (1-5)
mode: "alert-only"
# Severity (1-5, set by the AI on each verdict) at which "auto-action" automatically
# mutes the player instead of just warning them.
severity-threshold-mute: 4
# Severity at which "auto-action" automatically kicks the player instead of muting them.
# Should be >= severity-threshold-mute for the escalation to make sense.
severity-threshold-kick: 5
# Duration in minutes of the automatic mute applied by "auto-action" when crossing
# severity-threshold-mute (does not affect the manual /solver mute|tempmute).
mute-duration-minutes: 10
# Optional: force a specific sanction type for specific "situations" (AI tags), regardless of
# severity-threshold-mute/kick above - only used when mode is "auto-action". Valid tags:
# TOXIC, HARASSMENT, THREAT, HATE_SPEECH, SCAM, SPAM. Valid types: warn, mute, tempmute, kick,
# ban, tempban ('duration' required for tempmute/tempban, same s/m/h/d/w format as elsewhere).
# If a verdict matches more than one configured tag, the harshest configured type wins. This
# is also the only way to reach an automatic ban/tempban from chat moderation - the severity
# thresholds above never go past kick on their own.
tag-overrides: {}
# THREAT:
# type: tempban
# duration: 7d
# HATE_SPEECH:
# type: tempban
# duration: 3d
# SCAM:
# type: ban
# Multi-signal corroboration: the AI verdict alone can only trigger a real warn/mute/kick
# when it's at least this confident AND this severe. Below that, it also needs at least one
# corroborating pre-filter signal (caps/insult/repeated-target/burst - see pre-filter above);
# without one it stays as a staff-only alert for manual review via /solver check, never an
# automatic sanction. Confidence 0-100, severity 1-5.
min-confidence-to-act-alone: 85
min-severity-to-act-alone: 4
# Save every detected incident to moderation-reports/
log-incidents-to-file: true
# Also show in console the result of /solver moderation test <message> when a player
# runs it (the console always sees it already if it's the one running the command).
log-test-results-to-console: true
# How many days to keep incidents saved in moderation-reports/ before deleting them.
# 0 = keep forever.
reports-retention-days: 0
# Whether these command groups can also be run directly without the /solver prefix (e.g. /vanish
# instead of /solver vanish). /solver <command> always keeps working either way regardless of
# these settings. Turn a category off if another plugin you have installed already claims one of
# its command names (e.g. EssentialsX also has /mute, /kick, /ban, /warn, /tempban, /tempmute,
# /unmute, /unban) - Bukkit will otherwise let whichever plugin registers first keep the bare name.
# /enderchest is the one exception handled automatically: if EssentialsX is detected, its
# standalone form is skipped on its own (no need to toggle this whole category off for it) -
# EssentialsX's /enderchest means "show MY OWN enderchest", a different meaning from ours.
standalone-commands:
staff-mode: true # /vanish, /freeze, /staffchat, /commandspy, /enderchest (unless EssentialsX is installed)
sanctions: true # /warn, /mute, /tempmute, /kick, /ban, /tempban, /unban, /unmute, /unwarn, /history, /check, /note, /checkuser, /appeal, /sanctions
moderation: true # /moderation
reports: true # /report
# Live relay of executed commands to whoever has toggled it on with /solver commandspy - not just
# retrospective logging.
commandspy:
enabled: true
# "*" watches every command. Otherwise list specific command names (without the leading /) to
# watch only those, e.g. ["op", "gamemode", "give"].
watch-patterns: ["*"]
# Player names (case-insensitive) never relayed, regardless of who's watching.
exempt-players: []
# Anyone holding one of these permissions is never relayed either.
exempt-permissions: []
# ⚠ BETA: Sanctions System (warn/mute/kick/ban via /solver, persistent history in
# sanctions.db). Double-check /solver history after any sanction until you're fully
# comfortable with how it behaves on your server.
punishments:
# Where sanction history is stored: sqlite (default, zero config, local file) or mysql
# (for multi-server setups sharing one sanctions database).
storage: sqlite
# Only read when storage is mysql.
mysql:
host: localhost
port: 3306
database: solver
user: solver
password: ""
# Shortcut reason templates: "/solver ban Player #griefing" expands to the configured text
# (and duration, if the sanction type needs one and none was given explicitly). Unmatched
# #names are left as a literal reason, so this never breaks a reason that legitimately starts
# with "#".
reason-templates: {}
# griefing:
# text: "Griefing / destroying other players' builds"
# duration: "1d"
# Auto-escalation on warn count: after a WARN is recorded, if the player's total warn count
# matches a key here, the configured action fires automatically. "action: sanction" applies
# another sanction (type/reason/duration, same as a manual one); "action: command" runs a
# console command instead ({player} is replaced with the player's name).
warn-escalation: {}
# "3":
# action: sanction
# type: kick
# reason: "Auto: 3 warnings"
# "5":
# action: sanction
# type: tempban
# duration: "1d"
# reason: "Auto: 5 warnings"
appeals:
# Show a line pointing to /solver appeal <id> <reason> in warn/mute/kick/ban messages.
show-in-sanction-message: true
# ⚠ BETA: Staff Mode Toolkit (vanish/freeze/staff chat via /solver).
staff-mode:
vanish:
# Also subtract vanished players from the online-player count shown in the server list (MOTD)
hide-from-server-list: true
# Also make a vanished player immune to damage while vanished.
invulnerable-while-vanished: false
freeze:
# Commands allowed while a player is frozen (without the leading /)
allowed-commands: ["msg", "tell", "r", "helpop"]
# Named combined profiles for /solver staffmode <profile> - each toggles a bundle of the
# individual staff tools at once instead of turning them on one by one. Every field is optional
# and defaults to off. clear-inventory always saves your items first and restores them when you
# leave the profile.
profiles: {}
# moderator:
# vanish: true
# freeze-immunity: true
# flight: true
# god-mode: true
# clear-inventory: true
# builder:
# flight: true
# /solver inspect <player> - view a player's inventory without physically opening it.
inspect:
# Keep refreshing the opened inventory every few ticks instead of a one-time snapshot. This is
# polling, not a real push update (Bukkit has no "inventory changed" event to react to).
live-refresh: false
live-refresh-interval-ticks: 10
Commands
Base command: /solver (alias: /as). Most subcommands also work without the /solver prefix — see standalone-commands in config.yml.
Commands
/solver help: Displays the list of available commands.
/solver reload: Reloads the plugin's configuration and messages.
/solver analyze-last: Re-analyzes the last detected crash/error.
/solver crashme <exception|deadlock|oom|dry-run>: Triggers a synthetic error for testing the AI analysis pipeline.
/solver moderation status: Shows the current chat moderation diagnostics (buffer, last verdict, action mode, scan mode, AI analysis level, etc).
/solver moderation test <message>: Runs a message through the AI moderation prompt without it being a real chat message.
/solver warn <player> <reason>: Issues a warning. Reason accepts a #template from reason-templates.
/solver mute <player> <reason>: Mutes a player indefinitely.
/solver tempmute <player> <duration> <reason>: Mutes a player for a set duration.
/solver kick <player> <reason>: Kicks a player.
/solver ban <player> <reason>: Bans a player indefinitely. Also instantly kicks any known alt already online.
/solver tempban <player> <duration> <reason>: Bans a player for a set duration.
/solver unban <player>: Lifts an active ban/tempban.
/solver unmute <player>: Lifts an active mute/tempmute.
/solver unwarn <player>: Revokes the player's most recent active warning.
/solver history <player>: Shows a player's full sanction history.
/solver check <id>: Shows the details of a single sanction by its ID.
/solver note <player> <text>: Attaches a staff-only note to a player, without issuing a sanction.
/solver checkuser <player>: Cross-references IP history - shows every known alt account and any active sanction across that whole network.
/solver appeal <id> <reason>: Lets a sanctioned player appeal their own sanction in-game. Open to everyone by default.
/solver appeal list|accept|reject <id>: Staff review pending appeals.
/solver sanctions <player>: Opens a GUI browsing a player's full sanction history; click an active one for the exact revoke command.
/solver report <player> <reason>: Files a report against a player for staff attention. Open to everyone by default.
/solver reports: Lists open reports.
/solver reports claim|close|reopen <id> [reason]: Manages the reports queue. Notifies the original reporter.
/solver reports gui: Opens a GUI of open reports (left-click claims, right-click closes).
/solver vanish [strict]: Toggles your own vanish (hidden from other players, mob targeting, join/quit messages, tab-complete, and the server list count). "strict" is a second tier invisible to most staff too.
/solver freeze <player>: Toggles a freeze on a player (can't move, can't run most commands, can't break/place blocks, can't deal or take combat damage). Always toggleable even by the frozen player themselves.
/solver freeze <player> <message>: Sends a message in that player's private freeze chat channel instead of toggling.
/solver staffchat: Toggles staff-chat mode (your normal chat messages go to staff only instead of public chat).
/solver staffchat <message>: Sends a one-off message to staff chat without toggling the mode.
/solver commandspy: Toggles a live relay of other players' executed commands into your chat.
/solver staffmode <profile>: Toggles a named combined profile (vanish/freeze-immunity/flight/god-mode/inventory-clear) defined in config.yml.
/solver fly: Toggles flight, independent of any profile.
/solver god: Toggles invulnerability, including PvP damage.
/solver rtp: Teleports you to a random online non-staff player.
/solver inspect <player>: Opens a view of a player's inventory/armor/offhand without opening it; click an item to confiscate it.
/solver enderchest <player>: Same view/confiscate treatment for a player's ender chest.
Permissions
## Permissions
solver.admin: Full access to every command below. Default: op.
solver.notify: Receive error and chat-moderation staff alerts. Default: op.
solver.moderation.bypass: Exempts this specific player from chat moderation, regardless of OP status. Default: false.
solver.diagnostics.*: All diagnostics commands (reload/analyze-last/crashme/moderation). Default: false.
solver.diagnostics.reload: Allows /solver reload.
solver.diagnostics.analyzelast: Allows /solver analyze-last.
solver.diagnostics.crashme: Allows /solver crashme.
solver.diagnostics.moderation: Allows /solver moderation status|test.
solver.sanctions.*: All sanctions commands. Default: false.
solver.sanctions.warn: Allows /solver warn.
solver.sanctions.mute: Allows /solver mute.
solver.sanctions.tempmute: Allows /solver tempmute.
solver.sanctions.kick: Allows /solver kick.
solver.sanctions.ban: Allows /solver ban.
solver.sanctions.tempban: Allows /solver tempban.
solver.sanctions.unban: Allows /solver unban.
solver.sanctions.unmute: Allows /solver unmute.
solver.sanctions.unwarn: Allows /solver unwarn.
solver.sanctions.history: Allows /solver history.
solver.sanctions.check: Allows /solver check.
solver.sanctions.note: Allows /solver note.
solver.sanctions.checkuser: Allows /solver checkuser.
solver.sanctions.appeal: Allows /solver appeal <id> <reason> (submit your own). Default: true.
solver.sanctions.appeal.manage: Allows /solver appeal list|accept|reject.
solver.sanctions.gui: Allows /solver sanctions (history GUI).
solver.reports.*: All staff reports commands. Default: false.
solver.report: Allows /solver report (submit one). Default: true.
solver.reports.manage: Allows /solver reports listing/claim/close/reopen.
solver.reports.gui: Allows the reports GUI.
solver.staffmode.*: All Staff Mode Toolkit commands. Default: false.
solver.staffmode.vanish: Allows /solver vanish.
solver.staffmode.vanish.see-strict: See through another player's strict-tier vanish.
solver.staffmode.freeze: Allows /solver freeze.
solver.staffmode.staffchat: Allows /solver staffchat.
solver.staffmode.commandspy: Allows /solver commandspy.
solver.staffmode.profiles: Allows /solver staffmode <profile>.
solver.staffmode.fly: Allows /solver fly.
solver.staffmode.god: Allows /solver god.
solver.staffmode.rtp: Allows /solver rtp.
solver.staffmode.silentjoin: Auto-vanish on join with no join message.
solver.staffmode.inspect: Allows /solver inspect.
solver.staffmode.enderchest: Allows /solver enderchest.
Made with ♥️ by FyrxLab
评论 0
登录后参与评分
还没有评论,来抢沙发吧