Kivio Desktop v2.8.3

Kivio Desktop v2.8.3

桌面 AI 助手 · macOS (Apple Silicon) + Windows。Desktop AI assistant for macOS (Apple Silicon) and Windows.

下载 / Downloads

• Windows — Kivio.Desktop_2.8.3_x64-setup.exe（NSIS 安装包 / installer）
• macOS — Kivio.Desktop_2.8.3_aarch64.dmg（Apple Silicon，未签名 / unsigned）
  • 首次启动：右键 → 打开，或运行 xattr -cr "/Applications/Kivio Desktop.app"
  • First launch: right-click → Open, or run xattr -cr "/Applications/Kivio Desktop.app"

新版本亮点 / What's New

• 翻译 / 截图 / Lens 支持全部模型协议 / Translator, screenshot, and Lens now speak every model protocol — 这三个功能此前把请求硬编码成 OpenAI 的 chat/completions 格式，配置成 Gemini、Anthropic 或 OpenAI Responses 协议的供应商一用就 404。现在它们和聊天走同一套模型适配层，四种协议都能用于翻译、截图翻译和 Lens。These three features used to hard-code OpenAI's chat/completions shape, so any provider configured for Gemini, Anthropic, or OpenAI Responses returned a 404. They now share the same model adapter layer as chat, so all four protocols work for translation, screenshot translation, and Lens.
• 修复 Gemini / Vertex 与 grok 拒绝工具调用 / Fixed Gemini / Vertex and grok rejecting tool calls — 内置工具的 schema 含有分支未标注类型的 anyOf，Vertex 会整请求驳回、grok 会拒收工具定义，导致这些模型在 Kivio 里用不了工具。现已剥离该结构并由运行时校验兜底。A built-in tool schema carried an anyOf whose branches had no declared type; Vertex rejected the whole request and grok refused the tool definition, leaving those models unable to use tools in Kivio. The construct is now stripped, with runtime validation covering the constraint instead.
• 内置联网搜索 / Built-in web search — 每个会话可独立选择「关闭 / 内置 / 第三方」三态，内置模式直接调用供应商自带搜索（OpenAI Responses、Gemini、Anthropic），搜索过程以实时卡片显示在答案上方，不支持内置搜索的模型会置灰，最后一次选择记为新会话默认。Each conversation picks its own mode — off, built-in, or third-party. Built-in mode calls the provider's own hosted search (OpenAI Responses, Gemini, Anthropic) and streams it as a live card above the answer; models without hosted search are greyed out, and your last pick becomes the default for new conversations.
• 外部 CLI 全面重做 / External CLIs reworked — 全部 CLI 改用各自的原生会话机制，不再重放历史消息（kimi 迁到 ACP，pi 用 --session-id，会话与 CLI 绑定）；错误按类型分类并给出可操作的中文引导，原始错误折叠可展开；可用性探测与模型探测分离并加缓存，第 2 轮起前置开销从 10–25 秒降到 500 毫秒内；模型探测失败时降级显示并可重试。All CLIs now use their own native session mechanism instead of replaying history (kimi moved to ACP, pi uses --session-id, sessions bind to their CLI). Errors are classified with actionable guidance and the raw error folded away; availability and model probing are split and cached, cutting per-turn overhead from 10–25s to under 500ms from the second turn on; failed model probes degrade visibly and can be retried.
• Gemini 原生生图 / Native Gemini image generation — 聊天里可用 Gemini 原生协议出图并挂到 artifacts；OpenAI 兼容中转出图也已打通，出图端点路由收敛为单一解析并在猜错时自愈；模型只回文字不回图时也有兜底展示。Generate images in chat over Gemini's native protocol with results attached as artifacts; generation through OpenAI-compatible relays works too, with endpoint routing consolidated into one resolver that self-heals when it guesses wrong, plus a fallback when the model answers with text and no image.
• LaTeX 公式显示修复 / LaTeX rendering fix — 模型常用的 \[...\] 块级与 \(...\) 行内分隔符此前完全渲染不出来（#19），现已支持，代码块内不受影响。The \[...\] block and \(...\) inline delimiters that many models emit didn't render at all (#19); they now do, and code blocks are left alone.
• 从本地 CLI 导入 MCP 与 Skill / Import MCP servers and Skills from local CLIs — 扩展页可扫描已安装的 Claude Code / Codex / OpenCode 配置，勾选后导入其 MCP 服务器与 Skill；MCP 服务器改为在启用与开窗时预热连接，首轮工具收集加短超时兜底，一个慢服务器不再拖住整轮。The Extensions page scans your installed Claude Code / Codex / OpenCode configs and imports their MCP servers and Skills; MCP servers now warm up on enable and on opening the chat window, and the first tool listing has a short timeout so one slow server can't stall the turn.
• 独立截图标注 / Standalone screenshot annotation — 新增标注模式：箭头、矩形、马赛克，支持撤销后复制或保存；冻结帧双端常开；截图不再带鼠标指针；Windows 上二次截图会先隐藏浮窗，不再把 Lens 对话框截进去。A new annotation mode adds arrows, rectangles, and mosaic with undo, then copy or save; frozen-frame capture is on for both platforms, screenshots no longer include the mouse cursor, and on Windows a second capture hides the overlay first so the Lens dialog stays out of the shot.
• 新增 ego lite 插件 / New ego lite plugin — 插件中心新增 ego lite，启用时自动从仓库下载配套的 ego-browser Skill，随上游更新。The plugin center gains ego lite, which downloads its companion ego-browser Skill from the repo on enable and tracks upstream.
• 输入与界面修复 / Input and UI fixes — 所有自由文本输入框关闭 WebKit 智能输入（修首字母自动大写、中英切换时 autocorrect 插空格）；模型下拉按能力筛选（生图 / 视觉 / embedding 各只列具备该能力的模型，避免误选导致建索引失败）；弹层与右键菜单按视口边缘自适应，小窗口下不再截断；对话搜索可全文匹配消息正文；输入框草稿跨对话与页面切换保留；新增模型级 extra_body 透传。All free-text inputs disable WebKit smart input (fixing auto-capitalization and autocorrect inserting spaces when switching languages); model pickers filter by capability (image / vision / embedding each list only capable models, so you can't pick one that breaks indexing); popovers and context menus adapt to the viewport edge instead of being cut off in small windows; conversation search matches message bodies; composer drafts survive switching conversations and pages; and a new per-model extra_body passes custom request-body fields.
• 稳定性修复 / Stability fixes — HTTP 连接池加保活并缩短空闲淘汰（修长时运行后复用陈旧连接的 error sending request），网络错误展开完整原因链；用量统计的请求调试 token 不再恒为 0；修复切换/新建会话重复加载、多模型并答时流式栏冻结、Retry-After 巨值导致的长时挂起、后台命令未按会话隔离、知识库检索 rerank 分被覆盖与结果被静默截断。The HTTP connection pool gained keep-alive and shorter idle eviction (fixing error sending request from reusing stale connections after long runs), and network errors now show the full cause chain; request-debug token counts are no longer stuck at 0. Also fixed: duplicate loading when switching or creating conversations, the streaming column freezing during multi-model answers, long hangs from an oversized Retry-After, background commands not being isolated per conversation, and knowledge-base retrieval losing rerank scores or silently truncating results.

---

完整变更 / Full changelog: v2.8.2...v2.8.3