HanaAgent 1.5.118-beta

新功能与改进

• 草稿与附件跨重启保留：输入框附件会随草稿保存，只有图片的草稿也可以恢复；主页与 Code 页分别保留新会话草稿及附件。发送后仅移除本次已发送的附件。
• 发送前标注图片：在主输入框点击 PNG、JPEG、WebP 或 BMP 附件，即可使用画笔、圆角矩形、预设颜色、三档线宽和撤销。完成后替换待发送图片，取消或按 Esc 保留原图；GIF 和 SVG 仍为只读查看。
• 重启后恢复各页面会话：主窗口重新打开时，各页面会回到上次使用的会话；会话不存在时回退为草稿。
• 媒体设置更集中：同一供应商只显示一次，可集中管理凭据及图像、视频、语音和音频能力；支持多种接入方式的供应商可直接切换。
• 新增音频生成：支持豆包 Seed-Audio 1.0，可生成音效、环境音、带对白的场景音频和背景配乐；音频卡片显示格式与时长。
• 即梦支持页内登录：可在媒体设置中登录或退出即梦，并查看已登录、未登录、未安装或出错状态。
• 媒体生成前提供写法指南：首次使用某个供应商或模型时，助手会先读取对应指南，再提交生成。
• 减少会话切换开销：会话记忆总结改为空闲 10 分钟后执行，并降低后台检查与语音波形的资源占用。

问题修复

• 修复 Code 页并发加载时会话被误判为不存在、页面恢复和输入焦点异常，以及流式恢复重建时长会话短暂显示“加载中”的问题。
• 修复子代理结果已送达后仍重复唤醒、运行中的子代理无法接收补充消息的问题。
• 修复新对话记忆设置未跟随所选助手，以及会话回收或关闭时记忆总结收到空消息的问题。
• 修复损坏或无法读取的图片导致后续每轮请求失败的问题；不可用的图片现在会替换为占位文本。
• 修复窗口中可见的会话被缓存回收休眠的问题。
• 修复即梦已登录时误报错、等待期间临时网络错误中断登录，以及错误信息暴露设备码的问题；修正豆包音频时长显示。
• 修复卡片截图带入滚动条、自定义卡片显示根滚动条，以及着色会话行中 Code 分支名颜色不一致的问题。

升级须知

• 草稿图片副本共享 100 MB 本地存储。超过上限时，最久未更新草稿的图片副本会先被释放，文字草稿保留；清除草稿或删除会话会移除对应副本。
• 音频生成使用豆包语音供应商的凭据；即梦需要先安装 dreamina CLI，再在媒体设置中登录。
• 豆包语音的旧端点设置会迁移到新的“接入方式”；原“计费方式”区块由该选项取代。
• macOS 最低支持版本为 macOS 13。
• 本次只发布 Beta 通道；1.1.23-beta 之前的版本请使用完整安装包升级。
• Windows 安装包未签名，安装时可能出现未知发布者提示。

HanaAgent 1.5.118-beta

Features

• Drafts and attachments survive restarts: Composer attachments are saved with drafts, including image-only drafts. Home and Code keep separate new-session drafts and attachments. Sending removes only the attachments included in that message.
• Annotate images before sending: Click a PNG, JPEG, WebP, or BMP attachment in the main composer to use pen and rounded-rectangle tools, preset colors, three stroke widths, and undo. Done replaces the pending image; Cancel or Esc keeps the original. GIF and SVG remain read-only.
• Restore each page’s session after restart: The main window reopens the last session used on each page, falling back to a draft if that session no longer exists.
• More centralized media settings: Each provider appears once, with credentials and image, video, speech, and audio capabilities managed together. Providers with multiple connection options gain an access-method selector.
• Audio generation: Generate sound effects, ambience, scenes with dialogue, and background music with Doubao Seed-Audio 1.0. Audio cards show format and duration.
• Jimeng sign-in in settings: Sign in or out of Jimeng from media settings and see signed-in, signed-out, not-installed, or error status.
• Writing guides before media generation: Before the first generation with a provider or model, the agent reads its writing guide before submitting the request.
• Less session-switching overhead: Session memory summaries now run after 10 minutes of inactivity, while background checks and voice waveforms use fewer resources.

Bug Fixes

• Fixed overlapping Code page loads incorrectly treating sessions as missing, page restoration and input-focus issues, and long sessions briefly showing “Loading” during stream-resume rebuilds.
• Fixed redundant wake-ups after sub-agent results were delivered, and allowed running sub-agents to receive follow-up messages.
• Fixed new-chat memory settings not following the selected agent, and empty messages reaching memory summaries during session eviction or shutdown.
• Fixed damaged or unreadable images causing every subsequent turn to fail; unavailable images now become placeholder text.
• Fixed sessions visible in a window being hibernated by cache eviction.
• Fixed Jimeng reporting errors when already signed in, aborting sign-in after transient network errors, and exposing device codes in errors. Corrected Doubao audio duration metadata.
• Fixed scrollbars appearing in card screenshots and custom cards, plus inconsistent Code branch-name colors on color-tagged session rows.

Upgrade Notes

• Draft image copies share 100 MB of local storage. When the cap is exceeded, image copies from the least recently updated drafts are released first while their text remains. Clearing a draft or deleting a session removes its associated copies.
• Audio generation uses the Doubao speech provider’s credentials. Jimeng requires the dreamina CLI and sign-in from media settings.
• The legacy Doubao speech endpoint setting migrates to the new access-method selector, which replaces the former billing-mode section.
• macOS 13 or later is required.
• This release updates only the Beta channel. Clients earlier than 1.1.23-beta should upgrade with the full installer.
• Windows installers are unsigned and may display an unknown-publisher warning.

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.