中文Change log
[主要更新]
- 新增主动视觉功能（默认关闭）。猫娘会在主动搭话时获取屏幕信息，或在语音对话中时不时窥屏。
- 更加完善的自定义API支持。现在你可以使用自定义API来自由搭配服务商。你可以使用免费API搭配阿里的TTS语音。
- 新增基于循环神经网络的主动降噪系统，有助于提升语音识别表现。
[次要更新]
- 更新了阿里对话模型Qwen3-Omni-1201的默认参数，缓解复读问题。
- 文本模式下新增了一系列复读检测，触发时会自动清除复读对话并提醒用户对策。
- 新增插件系统（处于实验状态）。
- 语音克隆时可以选择参考语音的语言了。
- 使用Ogg Opus而非PCM传输TTS数据，显著降低带宽压力。
- 使免费/Step模型的文本模式下同样支持Web Search。
- Docker容器新增nginx反向代理。
[Bug修复]
- 修复了UI界面中一系列点击事件和状态重置问题。
- 当猫娘被打断时，不会再出现重复的记忆块。
- 麦克风静音检测不再计算语音系统准备期间的时间。
English Change Log
[Major Updates]
- Added Active Vision feature (disabled by default). The catgirl can capture screen information when initiating conversations, or occasionally peek at the screen during voice chats.
- Improved Custom API support. You can now freely mix and match service providers using custom APIs. For example, you can pair a free API with Alibaba’s TTS service.
- Introduced an active noise reduction system based on recurrent neural networks, helping to improve speech recognition performance.
[Minor Updates]
- Updated the default parameters for Alibaba’s dialog model Qwen3-Omni-1201, reducing repetition issues.
- Added a series of repetition-detection mechanisms in text mode. When triggered, repeated dialogue will be automatically cleared and users will be notified with suggested countermeasures.
- Added a plugin system (experimental).
- Voice cloning now allows selecting the language of the reference audio.
- Switched TTS data transmission from PCM to Ogg Opus, significantly reducing bandwidth usage.
- Enabled Web Search in text mode for free/Step models as well.
- Added an nginx reverse proxy to the Docker container.
[Bug Fixes]
- Fixed a number of click-event and state reset issues in the UI.
- Prevented duplicate memory blocks from appearing when the catgirl is interrupted.
- Microphone mute detection no longer counts the system’s voice preparation time.
What's Changed
- Docker封装，部署和启动流程的优化。Docker部署文档更新 by @WUHINS in #157
- fix: prevent click-through in advanced settings and improve UX by @Tonnodoubt in #158
- 修复免费版本 UI 显示和保存后状态显示错误 by @Chiya-mt in #159
- Added noise reduce funcionality with RNNoise. by @wehos in #168
- 添加可以根据屏幕内容主动搭话的功能 by @Chiya-mt in #160
- fix: 在加载新模型和刷新页面时关闭设置窗口 by @CN-QanYi in #165
- Add user-facing plugin subsystem (plugin server, registry, runtime state) by @wislap in #161
- Added repetition detection and notification. by @wehos in #170
- Support custom API. Screenshot utils unified usage. by @wehos in #171
- Refactor main_server. Adjust build profiles. by @wehos in #173
- 解决了自定义API设置视觉模型后模型选择错误的问题 by @Chiya-mt in #174
- feat(交互): 在拖拽模型时临时禁用按钮事件拦截 by @CN-Zephyr in #175
- feat(主动对话):基于标题搜索的主动对话功能 & 将热门内容改为首页推荐并更新相关逻辑 by @CN-QanYi in #176
- 新增语音聊天时如果开启了主动视觉则每十秒共享一次屏幕（1帧画面） by @Chiya-mt in #179
- feat(系统路由): 添加AI回复内容验证规则 by @CN-Zephyr in #181
- Enhance denoise and AGC quality for better VAD. by @wehos in #184
- 修复语音或屏幕共享按钮状态异常的问题 by @Chiya-mt in #189
Full Changelog: 0.5.5...v0.6.0