Skip to content

v0.2.0 - Significant features: MITM support & Online AI Model integration

Choose a tag to compare

View all tags
@Xe-Persistent Xe-Persistent released this 08 Jan 06:09
· 285 commits to master since this release
v0.2.0
466a8d1

简体中文 (Simplified Chinese)

本次更新重点引入了 MITM 流量分析模式和在线 AI 模型支持，并对系统核心进行了重要的性能优化。

✨ 新功能

  • MITM 中间人支持: 新增基于 mitmproxy 的通用流量拦截模式。您现在可以在任意浏览器或移动设备上进行对局，Akagi-NG 将在后台无感抓包并提供 AI 辅助。
  • 在线模型支持 (AkagiOT): 正式支持通过 AkagiOT 协议调用远程服务器进行 AI 推理，大幅降低了本地运行 Akagi-NG 的硬件配置门槛。

🚀 改进与优化

  • 通用立直模拟引擎: 引入了 ReplayEngine 来处理所有的立直前瞻（Lookahead）模拟。通过本地快速回放历史记录，不仅大幅提升了响应速度，还完美解决了在线模型调用受限（429 Rate Limit）的问题。
  • 架构重构: 对核心代码进行了语义化重命名（AkagiBot → StateTrackerBot 等），使得项目结构更加清晰，职责更明确。

🐛 问题修复

  • 版本号显示: 修复了在正式环境中，应用版本号错误显示为 "dev" 的问题。
  • 立直前瞻精度: 修复了立直 Lookahead 候选项置信度显示异常（如均为 33% 平均分布）的 Bug，现在能正确反映 AI 的真实置信度。
  • 模型加载: 修复了特定本地模型加载失败的问题。

English

This release brings significant flexibility improvements with the introduction of MITM support and Online AI integration, alongside critical performance optimizations.

✨ New Features

  • MITM Support: Introduced a Man-in-the-Middle mode powered by mitmproxy. You can now play on your preferred browser or mobile device, while Akagi-NG transparently intercepts and analyzes game traffic in the background.
  • Online Model (AkagiOT): Added support for the AkagiOT protocol, allowing the use of remote AI servers for inference. This enables high-performance AI analysis without demanding heavy local hardware resources.

🚀 Improvements

  • Universal Riichi Lookahead: All Riichi Lookahead simulations (for both local and online models) now use a continuous local ReplayEngine. This ensures instant feedback and eliminates "429 Rate Limit" errors when using online APIs.
  • Architecture Refactoring: Renamed core components (AkagiBot → StateTrackerBot, model.py → network.py) to better reflect their responsibilities and improve codebase maintainability.

🐛 Bug Fixes

  • Version Detection: Fixed an issue where the application version was incorrectly displayed as "dev" in PyInstaller-packaged builds.
  • Lookahead Precision: Resolved a bug where Riichi Lookahead candidates would display incorrect, uniform confidence scores (e.g., all 33%).
  • Model Loading: Fixed a bug where specific local models failed to load properly.

Full Changelog: v0.1.1...v0.2.0

Assets 3
Loading