1.5.1.1

1.5.1.1 🛠️ Sync Fixes, Model Controls & Enhanced Agent Logic

📁 Data & Sync Stability

• 🛡️ Race Condition Fix: Optimized the directory creation logic in AbstractJsonRepository to prevent concurrent write conflicts. This resolves a critical issue where chat history could fail to save when using sync services (e.g., Obsidian Sync or iCloud).
• 📝 Robust Titling: Strengthened the stability of the conversation auto-naming mechanism to ensure consistent and reliable session titles.

🤖 Agent & Model Refinements

• ⚖️ Intelligent Model Switching: Resolved conflicts between Agent-specific model settings and manual user selections in the UI, ensuring the correct model is always prioritized during Agent tasks.
• 🛠️ Tool Call Compatibility: Enhanced the robustness of tool calling across different LLM providers to ensure more reliable execution.
• 🎨 Elegant Parameter UI: Replaced hardcoded model parameters with a clean, native UI for Sampling Parameters, allowing for more flexible and visually integrated control over model behavior.

📐 Apply View & UX

• 🔍 Enhanced Navigation: Enabled the "Next/Previous Change" buttons even for single-diff instances, allowing for faster positioning and review of changes within the editor.

---

1.5.1.1 🛠️ 同步稳定性修复、模型控制优化与 Agent 逻辑增强

📁 数据与同步稳定性

• 🛡️ 修复并发竞态: 优化了 AbstractJsonRepository 的目录创建逻辑，防止在多端并发写入时出现冲突。彻底解决了同步盘（如 Obsidian Sync、iCloud）环境下聊天记录偶发无法写入的问题。
• 📝 命名机制加固: 增强了对话自动命名机制的稳定性，确保会话标题生成更加可靠。

🤖 Agent 与模型配置优化

• ⚖️ 模型切换逻辑优化: 解决了 Agent 助手预设模型与用户手动选择模型之间的冲突，确保 Agent 模式下模型调用的优先级与预期一致。
• 🛠️ 工具调用兼容性: 进一步强化了 Agent 工具调用的兼容性，提升了不同大模型下工具触发的成功率。
• 🎨 采样参数 UI 重构: 移除了模型侧硬编码的默认参数，新增了视觉更优雅的采样参数配置界面，支持更直观地调整模型生成偏好。

📐 Apply 视图与 UX 改进

• 🔍 导航体验优化: 在只有单处差异（Diff）的情况下也启用了“上下跳转”导航按钮，支持在编辑器中快速定位改动位置，提升审核效率。

1.5.1

1.5.1 🤖 Agent Mode: Think Smarter, Act Faster

🤖 The New Agent Framework

• ⚡ Agent Service Singleton: Decoupled Agent logic from the Chat View into a dedicated AgentService. The AI’s "thought process" now runs independently, ensuring a more stable and persistent lifecycle.
• 🔄 Seamless Mode Switching: Introduced a new roller selector alongside the classic accordion interaction, making it more intuitive to toggle between standard Chat and Agent modes.
• 🛠️ Tool & Assistant Management:
  • Added a dedicated Agent settings partition for unified tool naming and management.
  • Enhanced assistant settings with support for custom parameters (with types) and configurable context window size.
• 🎨 UI Polish: Added a new "Empty State" icon and descriptive copy for the chat sidebar to guide you when starting fresh.

🧩 Modular Context & Skills System

• 🏗️ Skills Core: Implemented the foundation for AI "Skills" with support for Always/Lazy loading modes and permission controls.
• 🪄 Skill Creator: Introduced a skill-creator to streamline the development of new capabilities. Built-in formatting constraints are now injected as a default skill.
• 📸 Snapshot Mechanism: Fully implemented a file snapshot mechanism to ensure the AI always references the correct version of your content during a conversation.
• 🧠 Optimized Context Flow:
  • Current page documents are now conversation-level context rather than message-level mentions, improving semantic stability.
  • Context is now appended to the end of requests to significantly improve LLM cache hit rates and reduce latency.
  • In Agent Mode, only the page title and path are sent by default to keep the prompt concise.

🛠️ Bug Fixes & Stability

• 🛡️ Robust History Loading: Fixed a crash when loading chat history if an @ mentioned file had been deleted.
• 📱 Android Compatibility: Fixed an issue where missing Buffer.byteLength caused message sending to fail on Android devices.
• 🖱️ Input Focus Fix: Resolved a bug where the cursor wouldn't appear when clicking the area below the title if the input contained @ mentions.
• 🔄 Request Optimization: Fixed a loop issue (#130) where session titles would trigger repeated requests, preventing accidental 429/999 errors and token waste.
• ✅ Apply Flow: Fixed cases where the "Accept All Changes" action would occasionally fail to execute.
• 🧹 Code Maintenance: Refactored style.css into modules for better maintainability and faster UI rendering.

---

1.5.1 🤖 Agent 模式：敏思，捷行

🤖 全新的 Agent 框架

• ⚡ AgentService 单例化：将 Agent 运行逻辑从 ChatView 中剥离，由插件级单例管理。即使切换视图，Agent 的生命周期也保持独立运行，更加稳定。
• 🔄 模式无缝切换：保留原有的手风琴交互，并引入了全新的滚筒选择器，让用户在普通对话与 Agent 模式之间切换更加直观顺滑。
• 🛠️ 工具与助手管理：
  • 新增 Agent 设置分区，统一了前端工具的命名规则与管理界面。
  • 优化助手设置，支持带类型的自定义参数以及上下文轮数的精确指定。
• 🎨 界面细节优化：为 Chat 侧边栏增加了全新的“空白态”图标与模式描述文案，新手引导更友好。

🧩 模块化上下文与 Skills 系统

• 🏗️ Skills 基础能力实现：支持技能的权限控制与加载模式（始终加载/按需加载），为插件的功能扩展打下基础。
• 🪄 Skill Creator：引入技能创建器优化系统设计，并将复杂的格式约束转化为内置的示例 Skill 全量注入。
• 📸 统一快照机制：全面引入文件快照机制，统一上下文传递逻辑，确保 AI 引用内容的实时性与准确性。
• 🧠 上下文传递优化：
  • 将当前页面文档从消息级 Mention 升级为对话级上下文，增强了语义稳定性。
  • 上下文内容改为追加到请求末尾，有效提升了大模型缓存（Cache）的命中率，响应更快。
  • 在 Agent 模式下默认仅传递页面标题与路径，精简上下文，聚焦任务目标。

🛠️ Bug 修复与稳定性增强

• 🛡️ 历史记录加载加固：修复了因 @ 文件 被删除导致的历史聊天加载异常，增强系统鲁棒性，避免启动报错。
• 📱 Android 端兼容性：修复了因缺失 Buffer.byteLength 导致的 Android 端发送消息报错问题。
• 🖱️ 输入框焦点修复：解决了输入框在包含引用标签时，点击特定区域无法激活光标的偶发问题。
• 🔄 请求频率控制：修复了会话标题生成的重复请求 bug (#130)，避免触发 API 的 429/999 限流，节省 Token 消耗。
• ✅ Apply 视图修正：修复了部分场景下“接受所有更改”按钮点击不生效的问题。
• 🧹 代码结构优化：对 style.css 进行了模块化改造，降低了维护难度并优化了样式加载性能。

1.5.0

1.5.0 ✨ New Apply View, Editor Flow Upgrades & Bug Fixes

📝 Apply View: Fully Redesigned Review & Apply Flow

• 🎯 Introduced a new Apply View with inline diff highlighting so you can see changes down to the exact modified characters.
• 📏 Changed block logic to split by line (one line break per block) to improve diff granularity and make decisions more precise.
• 📐 Added a right-side action rail and vertical markers for each paragraph to clarify where actions apply and to make clicking easier.
• ➖ Removed the old “merge both” option to simplify the flow to a clear accept or keep choice.
• 🎨 Reworked layout and hover behavior to better align with the editor’s native look and feel.
• 🫧 Updated the bottom toolbar to a centered floating pill that feels lighter and stays out of the main text flow.

💬 Cursor Chat, Quick Ask & Smart Space

• 🔧 Decoupled the Cursor Chat toggle from the “send selection to selection chat” behavior for more fine grained control.
• 📤 “Send selection to chat sidebar” is now the default option, preserving the original workflow.
• 🧑‍💻 Quick Ask assistant choice is now independent from the chat sidebar assistant, so each surface can use its own best suited helper.
• 🛡 Improved Quick Ask Edit Mode robustness and fixed cases where Gemini or DeepSeek reasoning output was incorrectly treated as an empty result.
• 🧵 Fixed an issue where Smart Space continuation only showed thinking and result after clicking back into the editor.

📎 Message Management & File Mentions

• ✏️ Added support to edit and delete AI replies in the chat sidebar, making conversation cleanup easier.
• 📂 Improved @ file behavior so the currently open file is pinned at the top of the selection list.
• 🧩 Fixed missing mention text in the sidebar when the current file was auto added.
• 🔢 Unified badge and mention counting logic so reference counts are consistent.
• 🧹 The chat sidebar no longer shows file badges and inline previews, keeping the message list clean while context is handled by mentions.
• 📏 When selecting mention text, the character count suffix is always preserved, making it easy to see how long the reference is.

🤖 Gemini Channel Fixes

• 🧱 Fixed an issue where system prompts were not always passed correctly to Gemini in some edge cases. Thanks to @magicskysword.
• ⏹ Fixed a bug where Stop Generation did not work reliably for Gemini models.

📁 Data Consistency & Concurrency

• 📡 JSON chat history writes are now serialized, avoiding concurrent adapter writes to the same file.
• 🧊 This resolves rare cases where the vault could freeze and cause Apply or save operations to fail.
• 🧬 Added a dedicated non reasoning or default model type, and hid reasoning intensity controls for these models to avoid confusion.

🎨 UI & Settings Details

• 🧷 Settings page top tabs now auto size based on label content, improving readability for longer labels.
• 🌐 Fixed cases where option text was cut off in some languages, improving localization rendering.

---

1.5.0 ✨ 全新的 Apply 视图、编辑体验升级与多项 Bug 修复

📝 Apply 视图全面升级：更清晰的差异对比与操作体验

• 🎯 全新 Apply 视图，支持行内差异高亮，可以精确看到被修改的字符和片段。
• 📏 分段逻辑改为 按行分块，每个回车就是一块，差异粒度更细，判断更直接。
• 📐 段落右侧新增 操作轨与竖条提示，可操作区域更清晰，点击操作更顺手。
• ➖ 移除了旧版的「合并两者」选项，流程简化为保留原文或应用修改两种选择。
• 🎨 样式与 hover 行为对齐 Obsidian 编辑器原生设计，视觉连续性更好。
• 🫧 底部操作栏改为 居中悬浮胶囊样式，层级更合理，也更不容易挡住正文内容。

💬 Cursor Chat、Quick Ask 与 Smart Space 优化

• 🔧 将 Cursor Chat 开关 与“选中发送到选区对话”的行为解耦，控制粒度更细。
• 📤 「选中发送到 chat 侧边栏」现在作为 默认行为，延续既有使用体验。
• 🧑‍💻 Quick Ask 与 chat 侧边栏的助手选择彻底解耦，两处可以分别选择更适配的助手。
• 🛡 提升 Quick Ask 编辑模式 的鲁棒性，修复 Gemini 与 DeepSeek 思考输出被误判为空结果 的问题。
• 🧵 修复 Smart Space 续写必须先点击编辑器，才显示思考过程和插入结果的 bug。

📎 消息管理与文件引用改进

• ✏️ 支持在侧边栏中 编辑与删除 AI 回复，方便整理历史对话。
• 📂 优化 @ 文件 的选择逻辑，当前打开的文件会自动置顶显示。
• 🧩 修复自动添加当前文件后，侧边栏中缺失对应 mention 文本的问题。
• 🔢 统一 徽章与 mention 文本的计数规则，引用数量统计更加一致。
• 🧹 Chat 侧边栏对话区域现在 不再显示文件徽章与预览内容，消息列表更干净，文件上下文由提及进行管理。
• 📏 选中 mention 文本时会 始终保留字数后缀，方便快速判断引用长度。

🤖 Gemini 渠道修复

• 🧱 修复部分场景下 系统提示词无法稳定传递到 Gemini 的问题，感谢 @magicskysword 的反馈。
• ⏹ 修复 Gemini 渠道 Stop Generation 无法正常生效 的问题。

📁 数据一致性与并发控制

• 📡 JSON 聊天历史写入改为 串行化处理，避免多个适配器并发写同一文件。
• 🧊 解决极端情况下 Vault 可能出现的 卡死与 Apply 或保存失效 问题。
• 🧬 新增 非推理 / 默认模型类型，并隐藏该类型下的推理强度选项，避免产生误导。

🎨 界面与设置细节优化

• 🧷 设置页面顶部标签改为 根据内容自适应宽度，长标题不再紧绷或被挤压。
• 🌐 修复某些语言下设置选项文本被截断的显示问题，多语言体验更完整。

1.4.13

1.4.13 Settings Layout Refresh, Chat Sidebar Redesign & Smart UX Refinements

🛠 Bug Fixes & General Improvements

• Improved compatibility for some providers in Obsidian that previously hit CORS or network errors. These calls now work reliably, although streaming is still not supported for them.
• Polished Sparkle styling, including a fix where completion length labels in the Sparkle sidebar were hard to read in light mode.
• New file or selection badges are now collapsed by default, which keeps the message area clean.
• Synced mention text and file badges, and fixed cursor positioning and the removal logic for auto-added current document references.
• Added a max length limit for @ names, improved badge layout and styles so long names do not break the UI.
• Moved tool and settings badges above the input area to avoid covering file badges, and refined their appearance.
• Improved the placement and styling of TAB completion advanced settings in the Sparkle sidebar.
• Fixed a bug where multiple tool call responses were merged incorrectly, which could cause LLM API failures. Thanks to @pistata112 for the report.
• Added more padding at the bottom of the message list, avoiding cases where the input area is hidden behind the stop button.
• Fixed an issue where switching files could cause duplicate input references or leftover blank text.
• Removed the automatic title length limit, so auto generated conversation titles are no longer forcibly truncated.
• Quick Ask now supports preceding and following context and annotates the cursor position. Added configurable context length limits (default 5000 / 2000).
• Fixed pasted image insertion so that it matches the position logic used by @ file mentions.
• Selecting text in the editor now immediately syncs the selection to the chat sidebar, and a bug was fixed where losing focus caused selection to fail.
• Updated chat view so file badges and tags only appear in the message where they were added, instead of being shown globally.

---

⚙ Settings Page: Tab Based Layout

• Completely reworked the settings UI to use a tabbed layout. Sections are now clearly separated and easier to scan, which reduces cognitive load.
• Unified tab styling so the visual language is consistent.
• Tweaked scrollbar and content styling in the settings view to improve readability and long section behavior.

---

💬 Chat Sidebar Visual Design & Interaction

• The bottom bar in the chat sidebar now follows the same collapse logic as Quick Ask. The model selector can be collapsed, freeing up space when not needed.

• Officially implemented reasoning strength controls in the chat sidebar, including better support for Claude reasoning models.

• Refined spacing between reasoning content and the main answer body for a clearer visual hierarchy.

• Added compact message bubbles and auto collapse / expand behavior for user messages in the sidebar chat history. This keeps long threads navigable.

  • Fixed a bug where, after editing a past message, pressing Enter could not resend it properly.

• Polished conversation info styling, making metadata more readable and better aligned with the new sidebar look and feel.

---

🚀 YOLO 1.5.x is coming soon, with a new Agent mode, smarter assistant structure, smoother multi turn interactions and stronger automation capabilities that reshape how AI works inside Obsidian.

---

1.4.13 设置内容结构优化、Chat 侧边栏样式升级，以及一系列 bug 修复与体验改进

🛠 Bug 修复与体验优化

• 排查并修复了 Obsidian 中调用部分服务商时出现的 CORS 或网络错误，这些渠道现在可以正常调用接口（暂不支持流式传输）。
• 进一步优化 Sparkle 的样式，并修复了在浅色模式下 Sparkle 侧边栏中补全长度标签因为字体颜色导致难以辨认的问题。
• 新添加的文件徽章和选区徽章现在会 默认收起，消息区域不再被大量徽章占满。
• 同步了 mention 文本与文件徽章 的逻辑，修复光标定位异常以及自动添加当前文档时的删除行为问题。
• 给 @ 提及名称增加了 最大长度限制，并优化徽章的显示逻辑与样式，让长名字也能保持合理布局。
• 将 工具和设置徽章 上移到输入框上方，避免遮挡文件徽章，同时对整体样式进行了细化调整。
• 优化 Sparkle 侧边栏中 TAB 补全高级设置 的位置与样式，结构更清晰。
• 修复了多个 tool calls 的响应在聚合时被错误合并导致 LLM 调用失败 的问题，感谢 @pistata112 的反馈。
• 在消息列表底部增加了 额外留白，避免后续输入区域被 stop 按钮遮挡。
• 修复切换文件后，输入框中 引用文本重复或残留空白 的问题。
• 去除了自动命名标题的 长度限制，自动生成的标题将完整保留。
• Quick Ask 现在支持 前后文上下文，并在上下文中标注 光标所在位置，同时新增 Quick Ask 上下文长度设置（默认 5000 和 2000）。
• 修复图片粘贴的插入位置，使其与 @ 引用的逻辑保持一致。
• 鼠标选中文本内容后会 即时同步到 chat 侧边栏，并修复了因用户切换焦点导致选区同步失败的问题。
• 调整 chat 视图中文件徽章与标签的显示逻辑，现在只会在 对应添加文件的那一条对话记录中显示 当时添加的文件。

---

⚙ 设置页面标签化重构

• 设置页面全面重构，采用 标签页样式 将不同分区拆分展示，结构更清晰，用户浏览和配置时的心理负担更小。
• 统一了 标签页的样式设计，视觉风格更加一致。
• 调整设置页内的 滚动条与内容区域样式，长内容的显示与滚动体验更加顺滑。

---

💬 Chat 侧边栏视图样式与交互优化

• Chat 侧边栏底部工具栏的收缩逻辑参考 Quick Ask 的交互设计，模型选择区域可以收缩，在空间有限的场景下更灵活。

• 正式实装 chat 侧边栏的 推理强度控制，并增加对 Claude 推理模型 的适配。

• 优化了 思维内容与正文之间的间距与层级，阅读体验更清晰。

• 为侧边栏聊天历史新增 紧凑型用户气泡样式，并支持自动收缩与展开，长消息不再占据整屏空间。

  • 修复了用户在编辑历史消息后，无法通过 Enter 重新发送消息的 bug。

• 对话信息区域的 info 样式 也进行了优化，元信息展示更直观。

---

🚀 敬请期待即将到来的 YOLO 1.5.x 版本，我们将正式引入全新的 Agent 模式，更智能的助手结构、更顺畅的多轮交互、更强的自动化能力，将全面升级 Obsidian 内的 AI 体验。

1.4.12

1.4.12 Quick Ask Polishing, Sidebar Sparkle Settings Overhaul, and Embedding/Completion Upgrades

🛠 Bug Fixes & Functional Refinements

• Embedding Compatibility Boost: Added logic to support multiple embeddings response formats, improving robustness when using third-party models.

• Codebase Alignment with Obsidian Review:

  • Updated copyright year.
  • Prefixed all style class names to avoid conflicts with core/other plugins.
  • Removed runtime CDN dependency; PGlite now loads from local only (apologies for the larger package size 😭).
  • Replaced type assertions with instanceof for safer runtime checks.
  • Language setting now defaults to Obsidian’s system language.

• Refactored main.ts: Separated out responsibilities—Selection Chat, Chat View navigation, RAG/MCP coordination, writing workflows, inline completions—all moved to dedicated modules. main.ts now cleanly handles only lifecycle and dispatch.

• Increased default RAG fallback threshold to 20,000 tokens, reducing false-positive errors when embedding config is absent.

• Adjusted localization module to strictly follow Obsidian language conventions.

---

⚡ Quick Ask UX Improvements

• Reimplemented Quick Ask as a ViewPlugin overlay instead of a document decoration—fixes issues with enter key misfocusing after asking a question.
• Fixed-width panel: The panel no longer stretches with longer replies.
• Made the drag bar taller and extended draggable zones—resizable from both left and right bottom corners.
• Resolved cursor misalignment in the input field.
• Redesigned the footer layout: left section now collapses adaptively, preserving space for action buttons.

---

🎛 Sparkle Sidebar Revamp: Smart Space, Quick Ask & TAB Completion

• Sidebar Sparkle section now includes three dedicated areas for Smart Space, Quick Ask, and TAB Completion—with fully editable options.
• Replaced dropdown UI with a new component to fix styling/interaction issues.
• Fixed toggle logic where disabled Smart Space options were still being shown.
• Removed hover border effects for cleaner UI.

---

⭾ TAB Completion Upgrades

• Added a preset length selector: users can now choose from Short / Medium / Long completions.

  • Optimized selector layout and added animated transitions for switching lengths.
  • Removed legacy ANSWER: trimming logic—streaming now directly uses raw output with improved system prompt design.

• Auto-triggering: completions can now be auto-triggered after typing stops.

• Better organization of the Sparkle TAB settings panel.

• Added a new “Accept Completion” command, enabling keybinding-based confirmations.

• Removed over-aggressive post-processing like trailing whitespace stripping and cutoff heuristics—raw output format is now fully preserved.

💬 This is likely the penultimate version before 1.5.x, which will introduce the new UI for the agent era.

---

1.4.12 Quick Ask 体验升级、Sparkle 分区全面配置化，以及嵌入与补全功能优化

🛠 Bug 修复与功能优化

• 嵌入兼容性提升：新增兼容多种 embeddings 响应结构的逻辑，增强模型适配能力与稳定性。

• 配合 Obsidian 审核反馈，修改核心结构：

  • 更新版权年份；
  • 所有样式类名添加插件前缀，防止覆盖核心/其他插件；
  • 移除运行时 CDN 加载机制，PGlite 现从本地加载（包体增大 😭）；
  • 类型判断改为使用 instanceof；
  • 语言设定默认跟随 Obsidian 系统语言。

• main.ts 模块化重构：Selection Chat、Chat 导航、RAG 协调、MCP 协调、写作逻辑、内联补全全部拆分为独立模块，main.ts 仅保留生命周期和调度逻辑。

• 将 RAG 默认触发阈值提升为 20000 tokens，降低误触报错概率。

• 语言模块调整为完全遵循 Obsidian 的语言切换机制。

---

⚡ Quick Ask 使用体验升级

• 将 Quick Ask 从 doc decoration 改为 ViewPlugin overlay，修复输入后回车导致焦点错乱/重载的问题。
• 回复不再撑宽浮窗，面板宽度保持统一。
• 拖动条高度提升，支持左右下角拖拽修改面板尺寸。
• 修复光标与提示文字错位问题。
• 底部栏布局优化，左侧可折叠，优先压缩模型/模式选择框，避免右侧按钮被挤出。

---

🎛 更好的侧边栏 Sparkle 分区配置（完全模块化！）

• Sparkle 侧边栏新增 Smart Space、Quick Ask、TAB 补全 三大分区，相关配置项集中管理。
• 引入新的下拉组件，解决原有下拉菜单的样式与交互限制。
• 修复 Smart Space 的设置选项在关闭时未正确隐藏的问题。
• 去除选项卡片 hover 时的主题色边框效果，界面更简洁。

---

⭾ TAB 补全功能优化

• 支持 预设补全长度：短 / 中 / 长 三档可选。

  • 补全长度选择界面重新布局，添加动画切换效果；
  • 移除旧版 ANSWER: 裁剪逻辑，直接使用模型原始流式输出，并升级系统提示词。

• 新增 自动触发补全 功能，输入停顿后自动唤起补全建议。

• 优化 Sparkle 面板中 TAB 配置区域的布局结构。

• 新增 Command 命令：“接受补全”，可绑定快捷键操作。

• 删除补全时的空白裁剪与内容边界清理，完全保留模型生成的原始格式。

💡 这可能是 1.5.x 到来前的倒数第二个版本，下一版本将正式推出面向 agent 时代的 全新 UI。

1.4.11

1.4.11 Performance Boosts, Chat History Overhaul & Stability Fixes

🚀 Performance & Stability Improvements

• Lazy Initialization: Chatview no longer auto-initializes DB/RAG/MCP providers on startup—they now load only when used, significantly reducing startup lag in Obsidian.

• Settings Behavior Polishing:

  • Numeric fields in settings no longer revert to default when left empty.
  • RAG section in settings is cleaner: long-tail configs are now collapsed under Advanced Options to reduce visual clutter.
  • Smart Space, Quick Ask, and Tab Completion settings are now shown in modal pop-ups, making the settings panel less crowded.

• New Commands Added: Quick commands for Smart Space, Quick Ask, and TAB Completion can now be triggered via the command palette or custom keybindings—even on mobile.

• Streaming Resilience: Introduced better error handling for streaming outputs in Smart Space and Quick Ask, improving reliability.

• Model-Specific Sampling Parameters: temperature and Top P are now defined per model, not globally. If left blank, they default to the model’s own settings, preventing misconfigurations.

💬 Chat History Refactor

• New Index File (chat_index): Introduced a dedicated index file to act as the single source of truth for chat history, improving robustness and data consistency.
• Simplified the logic for generating chat titles, making it more stable and predictable.
• Added Search Support for chat history, with improved dropdown width and styling.
• Introduced starred (favorite) chats, allowing important conversations to be pinned and prioritized.

---

1.4.11 启动性能优化与聊天记录系统重构

🚀 性能优化与稳定性提升

• 懒加载机制上线：Chatview 不再自动初始化 DB / RAG / MCP，改为按需加载，显著减少 Obsidian 启动时的卡顿现象。

• 设置项行为优化：

  • 数值项在清空时不再立即回退为默认值，用户体验更合理；
  • RAG 设置区的部分长尾配置项已移入 “高级设置”折叠块，界面更清爽；
  • Smart Space、Quick Ask 和 TAB 补全的设置项现在以 弹出式 Modal 展示，避免主设置页拥挤。

• 新增快捷命令：为 Smart Space、Quick Ask 和 TAB 补全添加 Obsidian Command 支持，可通过命令面板、移动端或快捷键快速触发。

• 流式传输容错增强：Smart Space 与 Quick Ask 的流式输出加入更稳健的错误处理机制，提升功能稳定性。

• 采样参数逻辑调整：temperature 与 Top P 现在改为 模型级参数，不再使用全局默认；若未填写则使用模型内建默认值，避免参数冲突。

💬 聊天记录系统重构

• 引入独立的 chat_index 索引文件，作为聊天记录的唯一来源，提高记录系统的 健壮性与一致性。
• 简化聊天标题生成逻辑，稳定性更高，命名更合理。
• 聊天记录列表新增 搜索功能，下拉框宽度与样式也同步优化。
• 支持对聊天记录添加 星标收藏功能，重要会话可置顶管理。

1.4.10

1.4.10 TAB Completion Refactor, Bug Fixes & Vector Indexing Performance Boost

🔁 TAB Completion Overhaul

• Rebuilt the entire TAB completion logic to deliver a smoother and more intuitive experience:

  • Adopted a “masked substitution” prompt strategy, enhancing model comprehension of local context.
  • Refactored how context is sliced and passed to the model to maximize relevance.
  • Consolidated settings to simplify configuration.
  • Multiple trigger phrases are now supported—flexible for different writing styles.
  • Streaming output is now available during TAB completion.
  • Fixed issues where completions failed to trigger or had redundant default triggers.
  • Added support for user-defined prompt rules specifically for TAB completions.

📘 A devlog “Designing TAB Completion in YOLO” is planned. README update coming soon.

---

🛠️ Bug Fixes & Performance Improvements

• Limited the use of the prediction parameter to OpenAI channels only, preventing failure on unsupported providers.

• Massively improved vector indexing performance for large note vaults:

  • Phase 1: Reduced 1000+ redundant queries to 1, optimized main-thread blocking, added async-safe save helpers, and improved UI responsiveness.
  • Phase 2: Introduced incremental saves, cancelable indexing, and cancel button UI.
  • Caveat: performance limitations still exist due to current reliance on PGlite, which may bottleneck at scale.

• Fixed numerous issues flagged by Obsidian’s plugin review.

• Corrected Italian translation errors (thanks @MarcoBarna).

• Sidebar settings position refined (thanks @kemroy3932).

• Smart Space quick actions now auto-select dropdown direction (prefer down, fallback up) and fixed scroll jitter & visual overflow at page bottom.

• Resolved layout conflicts where Smart Space, Cursor Chat, and Quick Ask could block window title bars.

• Adjusted PGlite shim regex to support Windows-style backslash paths, avoiding module resolution failures during RAG setup.

• Quick Ask context now includes the current page title, improving relevance.

• Refined Quick Ask option hover/active text color to reflect the theme and remain visible in light mode.

• Refactored and modularized main.ts, extracting YOLO-specific features.

• Applied updates to conform with latest Obsidian plugin review guidelines.

---

1.4.10 TAB 补全重构、Bug 修复与向量数据库性能优化

🔁 TAB 补全功能重构

• 对 TAB 补全功能进行了彻底重构，提供更自然、可靠的使用体验：

  • 引入 “掩码替代”结构的提示词设计，强化模型对上下文的理解；
  • 重新设计上下文截取逻辑，更精准地构造补全文本；
  • 合并精简设置项，配置更简洁；
  • 支持设置 多个触发词，适配不同写作习惯；
  • 补全过程支持 流式输出，响应更实时；
  • 修复无法触发补全的问题，并删减多余默认触发词；
  • 新增 自定义补全提示词规则，适配进阶用户需求。

📘 后续将发布开发日志《YOLO 的 TAB 补全设计》，并同步更新 README。

---

🛠️ BUG 修复与性能提升

• 仅为 OpenAI 渠道提供 prediction 参数，避免其他渠道误用后导致修改失败。

• 显著优化大规模笔记库的向量索引性能：

  • 阶段一：解决 N+1 查询问题，主线程释放、异步保存与工具函数改进，实现 1000 个文件仅 1 次查询，UI 不再冻结；
  • 阶段二：引入 增量保存、可中止索引、取消按钮 UI；
  • 注：随着文档库增长，PGlite 本身的瓶颈仍可能成为限制因素。

• 修复 Obsidian 插件审核反馈 中要求更改的内容。

• 修复 意大利语翻译错误（感谢 @MarcoBarna）。

• 调整 聊天侧边栏设置的位置（感谢 @kemroy3932）。

• Smart Space 快捷选项栏现支持 自动选择展开方向（优先向下，空间不足向上），并修复页面底部遮挡/滚动抖动问题。

• 修复 Smart Space、Cursor Chat、Quick Ask 可能遮挡 Obsidian 标题栏的问题。

• 将 pglite shim 的匹配正则改为兼容反斜杠路径，避免 Windows 系统下模块加载失败。

• Quick Ask 现在会将 “当前页面标题”加入上下文，增强回答相关性。

• Quick Ask 的选项框在 hover / 选中 / 展开状态下的文字颜色改为 主题色，防止浅色模式下看不清。

• 对 main.ts 进行模块化改造，抽离 YOLO 专属逻辑，结构更清晰。

• 根据 Obsidian 官方插件审核要求 更新代码结构与风格。

1.4.9

1.4.9 Italian Translation Added, UX Tweaks & Bug Fixes

🌍 New Language Support

• Added full Italian translation 🇮🇹 — special thanks to @MarcoBarna!

🎨 UX Improvements & Functional Fixes

• Fixed a bug where Selection Chat interfered with dragging the window title bar, improving layout fluidity.
• Smart Space trigger behavior on blank lines is now configurable with three preset modes.
• Fixed a bug where Quick Ask's post-edit feedback message failed to correctly display variable values.
• Resolved incorrect file list positioning when using @ in Quick Ask.
• Fixed an issue where files added via @ in Quick Ask were not properly included in the prompt context.
• The chat sidebar now supports aggregated display of all file references used in history, and deletes will now remove all associated references accordingly.

---

1.4.9 添加意大利语翻译，体验优化与 Bug 修复

🌍 新语言支持

• 新增完整的 意大利语翻译 🇮🇹 ——特别感谢 @MarcoBarna 的贡献！

🎨 使用体验优化与功能修复

• 修复 Selection Chat 可能阻碍窗口标题栏拖动 的问题，窗口交互更顺畅。
• Smart Space 在空行触发方式 现可自定义，支持三种预设选项，满足不同书写习惯。
• 修复 Quick Ask 编辑后提示信息中变量无法正常显示 的问题。
• 修复 Quick Ask 中使用 @ 添加文件时 文件列表定位异常 的问题。
• 修复 Quick Ask 中通过 @ 添加文件时 文件内容未被正确传入上下文 的 bug。
• 聊天侧边栏现支持 聚合显示历史消息中所有文件引用，在删除引用时将 同步移除所有关联项。

1.4.8

1.4.8 Quick Ask Bug Fixes, Edit Mode Expansion & Reasoning Chain Support

⚡ Quick Ask Stability & UX

• Fixed a bug where Quick Ask couldn't be closed in some cases.
• Resolved an issue where autoscroll during generation prevented users from viewing upper content.
• Fixed a bug where generated text couldn't be selected or copied.
• Added a "Clear Conversation" button to reset the Quick Ask session instantly.
• Polished styles for a cleaner look; compacted mode selector size & font for better clarity.

🛠️ Functional Enhancements

• Integrated reasoning chain rendering for supported inference models within Quick Ask.
• Expanded Edit Mode to support more actions such as Continue / Rewrite—not just single-shot edits.

🚨 New Mode: Edit (Full Access)

• Added a powerful new “Edit (Full Access)” mode: allows AI to directly modify file content without requiring user confirmation for each step—ideal for aggressive automation workflows.

💡 Note: Quick Ask is evolving rapidly and may eventually replace Smart Space as the unified assistant interface.

---

1.4.8 Quick Ask Bug 修复与体验优化，新增“编辑（完全访问）”模式

⚡ Quick Ask 稳定性与交互体验

• 修复 Quick Ask 无法关闭 的问题，面板行为更加稳定。
• 修复生成内容时页面会 自动滚动到底部 导致用户无法查看上方内容的 bug。
• 修复 生成文本无法被选中 的问题，文本交互体验提升。
• 增加 “清空对话”按钮，一键重置当前对话内容。
• 微调 Quick Ask 样式；模式选择器字体与尺寸优化为更紧凑布局。

🛠️ 功能增强

• Quick Ask 现在支持渲染 推理模型的思维链（reasoning chain），增强可解释性。
• 扩展“编辑”模式，新增支持 续写 / 改写 等场景，不再局限于单句修改。

🚨 新增模式：编辑（完全访问）

• 新增 “编辑（完全访问）”模式，AI 可在用户不再逐步确认的情况下，直接修改文件内容，适用于需要高度自动化的编辑流程。

💡 提示：Quick Ask 正在快速演化，未来有可能会成为 Smart Space 的全面替代方案。

1.4.7

1.4.7 Quick Ask Debut, File UX Upgrades & Sidebar Enhancements

📂 File Interaction Enhancements

• Right-clicking in the file list now allows you to add files or folders directly to a conversation as context.
• Added expand/collapse toggles to file previews in the chat sidebar, making long file lists easier to navigate.

⚡ Quick Ask (Ask Anything, Anywhere)

• Introduced Quick Ask, a new floating assistant panel designed for instant Q&A or on-the-fly rewriting, always accessible.
• Fully draggable and resizable, inspired by tools like Cursor’s cmd+k.
• Supports two modes: "Ask" (Q&A) and "Modify" (text rewriting).
• Refined UI & simplified interaction flow for minimal distraction and fast access.
• Persistent assistant selection: your chosen assistant is saved to settings.currentAssistantId and automatically used next time you open Quick Ask.
• Related settings have been added to the preferences panel for better control.

---

1.4.7 Quick Ask 上线、文件交互增强与侧边栏优化

📂 文件交互增强

• 支持在文件列表中 右键添加文件或文件夹 到当前对话中作为上下文，操作更高效。
• 聊天侧边栏的文件预览区域新增 展开 / 收起按钮，便于管理大量引用文件。

⚡ Quick Ask（快速提问，随时随地@来解答问题）

• 正式上线 Quick Ask 浮动面板，提供轻量化的助手调用体验，适用于 即时提问或文本修改 场景。
• 面板支持 自由拖动与缩放大小，操作体验对齐 Cursor 的 cmd+k。
• 提供两种模式：“提问”（Q&A）与 “改写”（文本修改），满足多样需求。
• 全面优化交互流程与 UI 设计，使其更轻量、更直观。
• 助手选择支持持久化：选中的助手会写入 settings.currentAssistantId，下次打开默认沿用上次选择。
• 同步完善了相关设置选项，可在偏好设置中进行管理与切换。