HanaAgent 1.5.7-beta

包含自 v1.3.6-beta 以来的更新。

Features · 新功能

• 新增「闲聊」：从桌面侧边栏进入，和主助手聊天，不必另建工作会话。支持文字、图片、文件、表情和可交互卡片，入口显示最近消息与未读数量。助手可以主动问候、按约定提醒你；在设置 → 和〈助手名〉闲聊中调整主动程度、安静时段和通知。
• 闲聊语音：配置语音识别后，按住输入框说话，松开转成文字，确认后再发送。助手也能用语音回复，语音条支持展开文字；闲聊设置支持搜索并选择声音模型、选择音色和调整语音回复频率，未单独选择声音时跟随多媒体设置。
• 闲聊风格与表情：选择说话风格并预览气泡样式，记录相处偏好；创建、导入、编辑和导出表情系列，助手也可以收藏表情。App 可提供说话风格，但这项权限不包含读取或发送闲聊消息。
• 把闲聊中的事情交给工作会话：助手可以查看有权访问的会话，经你确认后新建会话或向已有会话发消息；支持代码会话及独立工作树。在闲聊中单独发送 /reset 可重新开始上下文，停止当前回复并取消待处理确认，已有聊天记录保留。
• 更多语音服务：多媒体设置新增 OpenAI 和 OpenAI 兼容服务的语音合成支持。
• 文件夹移除：文件夹选择器和工作室列表支持移除文件夹记录，操作前会要求确认。不会删除磁盘文件，但引用该文件夹的对话将无法再定位它；助手的主文件夹不能移除。
• 代码字体选择：新增 LXGW Bright Code 和系统等宽字体。未设置过代码字体时默认使用 LXGW Bright Code，已有选择保持不变，更纱黑体仍可选。

Performance · 性能优化

• 多任务运行时的会话列表：合并连续的列表刷新请求，内容未变化的会话不再重复更新显示，减少多个助手同时工作时的界面开销。
• 带图长对话：复用未变化图片的处理结果，减少连续追问和多步任务中对历史图片的反复读取、解码与转换。
• 后台聊天页面：不再运行不可见页面的思考动画，减少同时保留多个聊天页面时不必要的动画开销。

Bug Fixes · 问题修复

• 修复多助手场景下，聊天截图可能使用其他助手的名称、头像或保存位置的问题。
• 修复关闭「无限上下文」后，大上下文模型仍提前压缩聊天记录的问题；现在会按模型实际上下文容量决定压缩时机。

升级提示：提供 macOS ARM64、Windows x64、Linux x64 安装包及 beta OTA，不含手机安装包。建议使用完整安装包升级，以获得 OTA 不包含的桌面应用更新。升级后，已有历史会话的主助手可能主动发出首次闲聊问候，可在闲聊设置中调整。闲聊采用自动审核，新建会话或向其他会话发消息仍会先征求确认，并非所有操作都会逐一弹窗。语音功能需配置相应服务，服务商可能收费。

HanaAgent 1.5.7-beta — English

Includes changes since v1.3.6-beta.

Features

• Casual Chat: Open it from the desktop sidebar to talk with your primary assistant without creating a work session. Share text, images, files, stickers and interactive cards, with the latest message and unread count shown at the entrance. Your assistant can reach out and give reminders as agreed. Adjust proactive contact, quiet hours and notifications in your assistant’s Casual Chat settings.
• Voice in Casual Chat: With speech recognition configured, hold the input field to dictate, release to convert speech to text, and send when ready. Your assistant can also reply with voice messages, with expandable transcripts. Search for a speech model, choose a voice and adjust reply frequency in Casual Chat settings, or follow the default voice settings under Multimedia.
• Speaking styles and stickers: Choose a speaking style and preview its bubbles, record conversation preferences, and create, import, edit or export sticker collections. Your assistant can also collect stickers. Apps can provide speaking styles; that permission does not include reading or sending Casual Chat messages.
• Move from conversation to work: Your assistant can view sessions it is allowed to access and, after your confirmation, create a session or send a message to an existing one. Code sessions and separate worktrees are supported. Send /reset on its own in Casual Chat to start with fresh context, stop the current reply and cancel pending confirmations; existing chat history is kept.
• More speech services: Multimedia settings now support speech synthesis through OpenAI and OpenAI-compatible providers.
• Remove folder entries: Remove folders from the folder picker and Studio lists after confirmation. Files on disk are not deleted, but conversations referencing the removed folder can no longer locate it. An assistant’s primary folder cannot be removed.
• Code font choices: Added LXGW Bright Code and the system monospace font. LXGW Bright Code is the default when no code font preference has been saved; existing choices are preserved and Sarasa remains available.

Performance

• Session lists during parallel work: Coalesced repeated list refresh requests and avoided updating unchanged conversation rows, reducing interface work while multiple assistants are active.
• Long conversations with images: Reused processing results for unchanged images, reducing repeated reads, decoding and conversions of historical images during follow-up questions and multi-step tasks.
• Background chat pages: Stopped running thinking animations on hidden pages, reducing unnecessary animation work when multiple chat pages remain open.

Bug Fixes

• Fixed chat screenshots potentially using another assistant’s name, avatar or save location when working with multiple assistants.
• Fixed large-context models compacting conversation history prematurely when Infinite Context was disabled. Compaction timing now follows the model’s actual context capacity in that mode.

Upgrade notes: Installers cover macOS ARM64, Windows x64 and Linux x64, alongside beta OTA; mobile installers are not included. A full installer update is recommended to receive desktop application changes that OTA does not include. After upgrading, a primary assistant with existing session history may send an initial Casual Chat greeting; adjust proactive contact in Casual Chat settings. Casual Chat uses automatic review: creating a session or messaging another session still asks for your confirmation, but not every action prompts individually. Voice features require the corresponding services to be configured; providers may charge for their use.

下载说明 / Download note

请下载本页提供的桌面安装包。GitHub 自动生成的 “Source code” 压缩包对应公开仓库的历史代码，并非本次 beta 安装包的源码。

Use the desktop installers attached to this release. GitHub-generated “Source code” archives contain the historical public repository snapshot, not the source for these beta binaries.