Skip to content

[v0.9.0] The Beginning of NekoVerse

Latest
Latest

Choose a tag to compare

View all tags
@wehos wehos released this 20 Aug 17:39
v0.9.0
435a033
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

0.9 大版本来啦！我们终于走到了这一步。属于每一个热爱猫娘的人的社区——喵宇宙，以及悠怡的新衣服「洛丽塔」！

[主要更新]

  • 喵宇宙社区：社区功能正式上线！日常对话中会自然掉落卡片，可以融合、铸造、收藏与分享。
  • 新衣服：洛丽塔：全新 Live2D 服装「洛丽塔」上线，并已设为默认皮肤。
  • PNGTuber正式上线：内置 YUI Origin / Sister / Lolita 三套形象，同时支持第三方分层工程导入。
  • 挂件模式，以及更生动的猫咪模式：猫娘可以从屏幕边缘探出半个身子；猫咪模式现在能感知窗口并与桌面内容发生更多交互。
  • 更丰富的主动搭话：新增贴吧、YouTube、小黑盒、Twitch、B 站等来源 + 个性化音乐推荐；未被回应时会自动收敛。
  • 直播插件与战雷插件上线：人猫同播 / 猫猫独播两种模式，接入 B 站、抖音、Twitch 弹幕；《战争雷霆》副驾驶。
  • 外部记忆导入：支持从 OpenClaw / Hermes 导入，带预估时间与实时进度。
    声纹识别与独立语音识别进入 Beta（独立 ASR 默认关闭）。

[次要更新]

  • QQ 插件全量重构：节点化流程、分层情景提示词、新前端；连接泛用化为 OneBot，新增识图、语音转录、文件读取。
  • 更多平台接入：微信插件、B 站私信/评论/@ 通知、网易云独立播放插件、QQ 音乐播放源。
  • 插件生态：插件技能、可回滚本地升级、自启增强、市场本地账号、插件 UI 实时推送通道、plugin-cli publish。
  • 记忆系统升级：群聊记忆按会话作用域隔离 + 发言人信赖度；核心记忆按主体分 Token 预算，召回改工具调用；记忆浏览器重做。
  • 语音与音频：豆包 TTS、OpenAI 兼容 TTS、vLLM-Omni 声音克隆、播放设备选择、语音优先人格预设。
  • 繁体中文全链路补齐（提示词/情绪/点歌/活动感知此前对繁中零命中），并加 CI 门禁防回退。
  • 凭证管理中心重构 + 图文 Cookie 教程；B 站私信扫码登录。
  • 伴学插件增强、大量 UI 翻新、性能优化（空闲降帧、启动导入 -37%、点歌台按需加载、召回热路径瘦身）。
  • 其他：MCP 自定义 Headers、拖文件进聊天框读取、VRM 语义动作与 VMC 输出、小游戏微调，以及大规模内部模块拆分。

[Major Updates]

  • NekoVerse — Community features are officially live! Everyday conversations now drop cards you can fuse, forge, collect, and share.
  • New Outfit: Lolita: The brand-new Live2D outfit "Lolita" is here, and it's now the default skin.
  • PNGTuber Officially Launches: Avatars now officially support PNGTuber, with YUI Origin / Sister / Lolita built in, plus imports of third-party layered projects.
  • Widget Mode, and a Livelier Cat Mode: The new "widget mode" lets her peek out from the edge of your screen with half her body showing. Cat mode now senses your windows, interacts with more of what's on your desktop, and finds its own fun there.
  • Richer Proactive Chat: New content sources including Tieba, YouTube, Xiaoheihe, Twitch, and Bilibili, plus personalized music recommendations. She now backs off automatically when her messages go unanswered.
  • Streaming Plugin and War Thunder Plugin: The streaming plugin supports both "co-stream" and "solo stream" modes and connects to Bilibili, Douyin, and Twitch chat. The "Cat Girl Copilot" plugin has her follow the state of the battle in War Thunder, calling things out and playing alongside you.
  • External Memory Import: Import memories from external agents such as OpenClaw and Hermes, with an estimated duration and live progress.
  • Voiceprint and Standalone Speech Recognition (Beta): She can tell who is speaking, and speech recognition can now be configured independently from the chat model (off by default).

[Minor Updates]

  • QQ Plugin Rebuilt: Node-based flow, layered situational prompts, a new frontend and onboarding; connectivity generalized to OneBot, with image understanding, voice transcription, file reading, and self-awareness added.
    More Platforms: New WeChat plugin, Bilibili DMs / comment replies / @ notifications, a standalone NetEase Cloud Music playback plugin, and QQ Music as a playback source.
  • Plugin Ecosystem: Plugin Skills, rollback-safe local upgrades, better autostart, local account support in the plugin market, a real-time push channel for plugin UIs, default config profiles, author info, and plugin-cli publish for developers.
  • Memory System: Group chat memory is now scoped per conversation, with speaker trust, display-name mapping, and a retraction endpoint; core memory gets a per-subject token budget, and recall moved to tool calls. The memory browser UI was rebuilt.
  • Voice and Audio: Added Doubao TTS, an OpenAI-compatible TTS interface, and vLLM-Omni voice cloning; playback device selection; new voice-first Neko presets.
  • Traditional Chinese Coverage: Prompts, emotion handling, song requests, and activity sensing previously had zero coverage in Traditional Chinese — all backfilled, with CI gates to prevent regressions.
  • Credentials Center Rebuilt: Centralized credential management with an illustrated cookie guide; Bilibili DMs support QR login.
    Study Companion: Knowledge-graph filtering and relation display, OCR fallback notices, and a batch of stability improvements.
  • UI Refresh: Model settings and voice controls, the jukebox, My Profile, character card settings, the custom API panel, chat fonts and the compact chat window — all reworked, with more complete dark mode coverage.
  • Performance: Idle-state throttling across Live2D, VRM, MMD, and PNGTuber; lazy loading of heavy SDKs on the startup path (-37% import time); the jukebox now loads on demand; big cuts to memory recall hot paths and footprint.
  • Desktop and Packaging: Screen sharing gains source filtering and remembered windows, Tauri capture compatibility, fuller Docker and Nginx reverse-proxy setup, and Steam playtime achievements now driven by Progress Stats.
  • Others: Custom headers for MCP, drag-and-drop file reading in the chat box, VRM semantic motions and VMC motion output, smarter opening positioning in the soccer mini-game, an ice-block item in badminton, plus a large internal refactor splitting oversized modules.

What's Changed

  • PR2046-PR2925. That's huge!

New Contributors

Full Changelog: v0.8.3...v0.9.0

Contributors

fjfanjian, for-the-zero, and 8 other contributors
Assets 3
Loading