08 Jun 08:02

yl985211

Immutable

v3.0.0

b57af0a

GG-AI Modifier v3.0.0 Latest

Latest

GG-AI Modifier v3.0.0 版本更新说明 (Release Notes)

The English version is at the bottom of the document

在这个里程碑式的 v3.0.0 版本中，我们迎来了底层架构的彻底重构！为了追求极致的检索性能，我们将核心内存搜索引擎使用 C++ 原生重写。同时，针对 AI 自动化控制链进行了深度修复与对齐，并引入了全新的全局视觉主题。

🚀 一、核心性能飞跃：C++ 原生搜索引擎重构

我们彻底重写了内存搜索的底层逻辑。得益于 C++ 的并发优化与高效率内存指针调用，实际测试中的搜索响应速度实现了断层式飞跃：

小规模内存扫描 (200MB)： 仅需 0.53 秒 即可出结果。
大型游戏全段扫描 (5GB)： 仅需 4.5 秒 极速完成。
极大降低了大段内存扫描时的手机发热量，并有效避免了系统因大内存占用导致的 OOM（内存溢出）奔溃。

🎨 二、全局视觉与细节体验优化

全新淡黄主题： 引入全局淡黄色（Soft Cream）护眼主题，重构视觉规范，显著降低长时间调试或进行游戏内存分析时的眼睛疲劳度。
新增「插件中心」预留入口： 应用内新增了插件中心模块（当前版本暂无实际功能），为后续支持第三方自动化工具、高级逆向扩展奠定了框架基础。

🛠️ 三、核心 Bug 修复与优化

修复 AI 对话无法操作内存的 Bug： 彻底打通了 LLM 与底层引擎的壁垒。现在，AI 通过大模型函数调用（Function Calling）可以顺畅地对内存执行精确搜索、范围搜索、特征码匹配及读写操作。
修复模型切换异常： 解决了在不同大模型 API（如 DeepSeek、OpenAI 等）之间切换时可能导致的配置失效或不生效问题。
修复附加进程卡死： 优化了多线程锁机制，彻底解决了旧版本在目标游戏启动时进行进程附加可能导致的界面永久冻结、卡死现象。
精细化系统提示词（System Prompt）： 优化了 AI 的内置提示词模板，使其在理解游戏修改、内存偏移、特征码逆向等特定专业语境时的表现更加精准。

📊 四、底层引擎与 AI 交互能力适配矩阵

为了保证“手工面板操作”与“AI 自动化对话修改”的能力对齐，当前的适配与实现进度如下。C++ 原生引擎与 AI 工具箱（Function Calling）已保持高度一致：

操作类型	MemoryEngine 底层方法	当前 AI 工具支持 (Function Calling)	实现状态
精确搜索	`searchExact`	`search_memory(exact)`	已实现
范围搜索	`searchByRange`	`search_memory(range)`	已实现
AOB 特征码	`searchAob`	`search_memory(aob)`	已实现
模糊搜索	`searchFuzzy`	-	❌ 暂未实现
结果过滤	`filterResults`	-	❌ 暂未实现(有时候生效)
读取内存	`readMemory`	`read_memory`	已实现
写入内存	`writeMemory`	`write_memory`	已实现
收藏地址	`saveFavorite`	-	❌ 暂未实现
冻结数值	`toggleBookmarkFreeze`	-	❌ 暂未实现

注：未实现的功能（如模糊搜索、结果过滤、数值冻结等）已进入下一阶段的排期，UI 面板与 AI 控制端将同步在后续版本中更新上线。

感谢你对 GG-AI Modifier 项目的支持！此版本改动涉及较多底层重构，如有任何运行异常，欢迎在 Issues 中反馈。

GG-AI Modifier v3.0.0 Release Notes

In this milestone v3.0.0 release, we've completely rebuilt the underlying architecture! To achieve ultimate search performance, we've rewritten the core memory search engine using native C++. Simultaneously, we've performed deep repairs and alignment of the AI automation control chain and introduced a brand-new global visual theme.

🚀 I. Core Performance Leap: Rebuilt Native C++ Search Engine

We've completely rewritten the underlying logic of memory search. Thanks to C++'s concurrency optimizations and efficient memory pointer calls, the search response speed in actual testing has achieved a dramatic leap:

Small-scale memory scan (200MB): Results in just 0.53 seconds.
Full-segment scan of a large game (5GB): Completed in a lightning-fast 4.5 seconds.
Significantly reduces phone heat generation during large memory scans and effectively prevents system crashes due to OOM (Out of Memory) caused by large memory usage.

🎨 II. Global Visual and Detail Experience Optimization

New Pale Yellow Theme: Introduces a global pale yellow (Soft Cream) eye-friendly theme, reconstructing visual standards and significantly reducing eye fatigue during prolonged debugging or game memory analysis.
Added "Plugin Center" Entry Point: A plugin center module has been added within the application (currently without actual functionality), laying the framework foundation for future support of third-party automation tools and advanced reverse engineering extensions.

🛠️ III. Core Bug Fixes and Optimizations

Fixed a bug where AI dialogue could not manipulate memory: Completely breaks down the barriers between LLM and the underlying engine. Now, AI can smoothly perform precise searches, range searches, feature code matching, and read/write operations on memory through large model function calls.
Fixed Model Switching Issues: Resolved an issue that could cause configuration failures or ineffectiveness when switching between different large model APIs (such as DeepSeek, OpenAI, etc.).
Fixed Attached Process Freezing: Optimized the multi-threaded locking mechanism, completely resolving the permanent interface freezing or crashing issue that could occur when attaching processes during target game startup in older versions.
Refined System Prompt: Optimized the AI's built-in prompt templates, making them more accurate in understanding specific professional contexts such as game modifications, memory offsets, and feature code reverse engineering.

📊 IV. Underlying Engine and AI Interaction Capability Adaptation Matrix

To ensure alignment between "manual panel operation" and "AI automated dialogue modification" capabilities, the current adaptation and implementation progress is as follows. The C++ native engine and AI toolkit (Function Calling) are highly consistent:

| -------------- | ------------------------- | --------------------------------------- | ---------------------- |

Note: Unimplemented features (such as fuzzy search, result filtering, value freezing, etc.) are in the next phase of the schedule. The UI panel and AI control panel will be updated and launched in subsequent versions.

Thank you for your support of the GG-AI Modifier project! This version involves many underlying refactorings. If you encounter any runtime anomalies, please report them in the Issues section.

Assets 4

0 Join discussion

09 May 19:53

yl985211

Immutable

v2.0.0

630faaa

GG-AI Modifier v2.0.0

GG-AI Modifier 更新日志

English document at the bottom

v2.0.0 (2026-05-09)

🆕 新增功能

核心引擎升级

内存段智能筛选：解析 /proc/[pid]/maps 时自动按优先级排序，优先搜索 [heap]、[anon:*] 等高价值内存段，跳过只读段，提升搜索效率
模糊搜索（未知值搜索）：支持 8 种对比操作（changed/unchanged/increased/decreased/greater/less/equal/not_equal），首次搜索保存快照，后续对比变化
特征码搜索（AOB Scan）：支持字节模式匹配，记录目标地址前后 16 字节切片，实现游戏重启后自动重定位
数据对齐修复：使用 ByteBuffer.order(ByteOrder.LITTLE_ENDIAN) 显式声明字节序，支持 1 字节精细扫描步进
分块读取：内存扫描采用 2MB chunk 分块读取，避免 OOM

Lua 脚本引擎

集成 LuaJ (luaj-jse-3.0.2)：在 JVM 中执行 Lua 脚本，无需设备安装 luajit
完整 GG API 桥接：支持 gg.choice()、gg.prompt()、gg.toast()、gg.alert() 等交互式 API
交互式对话框：Lua 脚本中的 gg.choice() 会弹出真正的选择对话框，阻塞等待用户选择后继续执行（与 GG 修改器体验一致）
脚本运行日志：每次运行脚本自动保存日志，支持查看历史运行记录
脚本持久化：用户创建的脚本保存到 Hive 存储，重启后不丢失
悬浮窗脚本库：悬浮窗中可直接查看脚本列表并运行，日志与主应用同步

AI 对话升级

Markdown/LaTeX/Mermaid 渲染：悬浮窗 AI 对话使用 WebView 渲染，支持 Markdown 格式、LaTeX 公式（KaTeX）、Mermaid 图表
Flutter 端 Markdown 渲染：对话详情页使用 flutter_markdown 包渲染 AI 回复
真实 LLM API 调用：悬浮窗 AI 对话改为调用真实 API，支持多轮对话上下文

对话记录管理

单条/多条导出：支持导出单条或批量对话记录为 Markdown 文件
导出路径：/storage/emulated/0/Documents/AI-gg/（Android 10+ 使用 MediaStore API）
刷新按钮：悬浮窗保存聊天后可手动刷新，无需重启应用
真正删除：清空所有记录时同时清除 Hive 存储和 SharedPreferences

LLM 配置增强

自定义模型：支持添加自定义 API 提供商，填写名称/URL/Key/Model 后保存为标签
预设独立 API Key：每个预设独立保存 API Key，切换预设不会丢失已配置的 Key
预设持久化：重启应用后自动恢复上次选择的预设

设置页面

AI 读取深度：阶梯式采样开关（20/30/50/100/全部），限制发送给 AI 的搜索结果数量
GitHub 开源地址：点击跳转浏览器打开项目页面
首次启动权限申请：自动申请存储权限和悬浮窗权限

悬浮窗优化

记住上次面板：点击悬浮球恢复上次关闭的面板
每级窗口"关闭窗口"：所有面板底部都有"关闭窗口"按钮
悬浮窗脚本库：简化为只显示脚本列表和运行按钮
切换进程提示：AI 对话欢迎消息提示"切换进程后记得清空聊天"

🐛 修复

修复 Android 15 导出对话记录失败的问题（改用 MediaStore API）
修复悬浮窗 AI 对话硬编码回复的问题
修复对话内容切换后消失的问题
修复清除所有记录是伪删除的问题
修复 LLM 配置预设重启后变成"自定义"的问题
修复切换预设时 API Key 消失的问题
修复 Release 构建 R8 混淆器移除 LuaJ 类的问题
修复悬浮窗脚本库重复显示内置脚本的问题
修复悬浮窗脚本库打开新任务界面的问题

📦 版本信息

版本号：2.0.0+2
最低 Android 版本：Android 5.0 (API 21)
目标 Android 版本：Android 15 (API 35)

v1.0.0 (2026-05-07)

初始版本

基础内存搜索（精确搜索、范围搜索）
内存读写
内存冻结
进程管理（附加/分离）
AI 对话（LLM Function Calling）
Lua 脚本库
悬浮窗（悬浮球 + 功能面板）
设置页面（LLM API 配置、Root 权限检测）
首次启动权限申请

GG-AI Modifier Changelog

v2.0.0 (2026-05-09)

🆕 New Features

Core Engine Upgrade

Smart Memory Segment Filtering: Automatically sorts memory segments by priority when parsing /proc/[pid]/maps. High-value segments like [heap] and [anon:*] are searched first, while read-only segments are skipped, improving search efficiency.
Fuzzy Search (Unknown Value Search): Supports 8 comparison operations (changed/unchanged/increased/decreased/greater/less/equal/not_equal). The initial search saves a snapshot, and subsequent searches compare against it.
Signature Scan (AOB Scan): Supports byte pattern matching. Records a 16-byte slice before and after the target address, enabling automatic relocation after game restart.
Data Alignment Fix: Uses ByteBuffer.order(ByteOrder.LITTLE_ENDIAN) to explicitly declare byte order, supporting fine-grained scanning with a 1-byte step.
Chunked Reading: Memory scanning uses 2MB chunked reading to avoid out-of-memory errors.

Lua Script Engine

Integrated LuaJ (luaj-jse-3.0.2): Executes Lua scripts inside the JVM without requiring the device to have LuaJIT installed.
Complete GG API Bridge: Supports interactive APIs such as gg.choice(), gg.prompt(), gg.toast(), and gg.alert().
Interactive Dialogs: gg.choice() inside Lua scripts displays a real selection dialog that blocks and waits for user choice before continuing (matching the experience of GG Modifier).
Script Run Logs: Automatically saves a log for each script run, supporting viewing of historical run records.
Script Persistence: Scripts created by users are saved to Hive storage and are not lost after restart.
Floating Window Script Library: The floating window can directly display the script list and run scripts. Logs are synchronized with the main app.

AI Conversation Upgrade

Markdown/LaTeX/Mermaid Rendering: The AI conversation in the floating window uses a WebView for rendering, supporting Markdown, LaTeX formulas (KaTeX), and Mermaid diagrams.
Flutter-side Markdown Rendering: The conversation detail page uses the flutter_markdown package to render AI replies.
Real LLM API Calls: The floating window AI dialogue now calls a real API, supporting multi-turn conversation context.

Chat Record Management

Single/Batch Export: Supports exporting single or batch conversation records as Markdown files.
Export Path: /storage/emulated/0/Documents/AI-gg/ (Uses MediaStore API on Android 10+).
Refresh Button: After saving a chat in the floating window, you can manually refresh without restarting the app.
Real Deletion: Clearing all records now removes data from both Hive storage and SharedPreferences.

LLM Configuration Enhancements

Custom Model: Supports adding custom API providers. Fill in Name/URL/Key/Model and save as a new preset tag.
Per-Preset Independent API Key: Each preset saves its own API Key; switching presets will not lose the configured Key.
Preset Persistence: The app automatically restores the last selected preset after restart.

Settings Page

AI Read Depth: Stepped sampling toggle (20/30/50/100/All), limiting the number of search results sent to the AI.
GitHub Open Source Link: Tap to open the project page in a browser.
First-Launch Permission Request: Automatically requests storage and overlay permissions on first start.

Floating Window Optimization

Remember Last Panel: Tapping the floating ball restores the last closed panel.
"Close Window" on Every Level: All panels have a "Close Window" button at the bottom.
Floating Window Script Library: Simplified to show only the script list and run button.
Process Switch Hint: The AI conversation welcome message reminds users to "clear the chat after switching processes".

🐛 Fixes

Fixed export failure on Android 15 (switched to MediaStore API).
Fixed hardcoded replies in the floating window AI conversation.
Fixed conversation content disappearing after switching.
Fixed pseudo-deletion when clearing all records.
Fixed LLM configuration preset reverting to "Custom" after restart.
Fixed API Key disappearing when switching presets.
Fixed R8 obfuscator removing LuaJ classes in Release builds.
Fixed duplicate built-in scripts in the floating window script library.
Fixed the floating window script library opening a new task interface.

📦 Version Info

Version: 2.0.0+2
Minimum Android Version: Android 5.0 (API 21)
Target Android Version: Android 15 (API 35)

v1.0.0 (2026-05-07)

Initial Release

Basic memory search (exact search, range search)
Memory read/write
Memory freezing
Process management (attach/detach)
AI conversation (LLM Function Calling)
Lua script library
Floating window (floating ball + function panel)
Settings page (LLM API configuration, root permission detection)
First-launch permission request

Assets 4

1 Join discussion

06 May 11:36

yl985211

Immutable

v1.0.0

630faaa

GG-AI Modifier v1.0.0

功能特性

基于 Flutter + Kotlin 开发
LLM 大语言模型驱动游戏内存修改器

使用说明

下载 APK 文件
安装到 Android 设备
授予必要权限
启动应用开始使用

Assets 4

0 Join discussion

Releases: yl985211/gg-ai-modifier