RPClient 2026.4.2

📦 RPClient 2026.4.1

中文版本 (Chinese)

✨ 新增功能

• 主题配色与自定义背景：
  • 新增经典蓝、樱花粉、森林绿、暮光紫、暖纸棕、石墨灰六组配色，并支持系统动态配色。
  • 支持为通用页面、单聊、群聊和故事页面设置背景图片，可调整图片位置、模糊度与背景遮罩。
  • 菜单栏、底部操作栏和内容区可分别设置透明度与背景模糊度；单聊、群聊和故事页面可选择跟随通用设置或独立调整，并提供效果预览。
  • 角色卡支持设置专属背景，单聊可选择优先使用角色背景。
  • 新增对话框外观设置，可调整对话框背景的透明度与模糊度、内部控件背景的透明度，以及后方页面的压暗和模糊效果。
• 历史内容搜索与角色筛选：
  • 首页新增内容搜索入口，支持按单聊、群聊、故事分类，并选择搜索全部内容、标题与角色或正文。
  • 正文搜索覆盖全部已保存的聊天消息和故事章节，可从结果跳转至对应历史消息或章节，不受当前已加载消息和当前章节限制。
  • 创建单聊、群聊和故事时，可按角色名称、描述或标签筛选角色，保留已有选择。
  • 新增可选的搜索加速功能，手动开启并完成索引构建后生效；支持取消构建，以及关闭后清理索引、回收空间。
• 完整数据备份与恢复：（@onemind66）
  • 新增“备份与恢复”页面，可将角色、单聊、群聊、故事、世界书、正则脚本、模型配置、偏好设置和附件等应用数据打包导出为 ZIP。
  • 备份包含 API 密钥、Token 用量和已保存的请求日志，数据按功能组织为 JSON / JSONL，便于查看与保存。
  • 支持从备份恢复数据，恢复前检查归档完整性与数据库兼容性，完成后自动重启应用。
• 模型协议与会话控制：
  • 新增独立的 OpenAI Responses 原生协议，支持文本与图片输入、流式回复、推理内容解析和服务端用量统计。
  • 新建 OpenAI 模板默认使用 Responses 协议，已有模型配置保留原协议，也可在高级设置中手动切换。
  • 模型配置新增模型级 System Prompt，使用该模型的请求会前置添加专属指令，适用于聊天、故事生成、摘要和连接测试。
  • 单聊与群聊新增“注入角色卡示例对话”开关，可在创建时或会话设置中单独控制，并继续遵循全局示例对话策略。
• 故事阅读与消息编辑：
  • 故事章节支持主题化 Markdown 预览，可切换回原文编辑，并提供快速滚动到顶部、底部的操作。（@onemind66）
  • 完善故事搜索结果的高亮与定位，方便从搜索结果返回原文继续编辑。（@onemind66）
  • 新增“分离思考链”设置，默认在编辑消息时仅修改正文、保留原有思考链；关闭后可编辑完整消息。
• 调试日志查看与导出：
  • JSON 查看器支持“可视化”和“源码”两种视图，方便检查已保存的请求与响应内容。
  • 支持将查看的内容导出为格式化 JSON；无法格式化时，可按已保存的原文导出。
  • 流式响应日志改为按事件组织，保留事件信息与响应载荷，便于排查流式请求问题。

🎨 体验优化

• 界面与操作一致性：
  • 统一单聊、群聊、故事、世界书和正则编辑等页面的标签栏样式，优化设置、模型配置、备份与恢复及各类创建页面的排版。
  • 统一世界书选择与管理组件、创建页操作栏和按钮的透明与模糊效果，改善浅色模式下的视觉表现。
  • 精简部分重复说明，调整分支创建图标，并扩大设置页模型配置管理入口的点击区域。
• 参数与运行开销：
  • Token 快捷选项和模型预设预算统一采用十进制数值，方便直接理解和调整额度。
  • 关闭应用内调试模式时，跳过请求日志采集、正文构建和保存相关处理，减少额外开销。

🐛 问题修复

• 修复群聊消息未正确显示对应角色头像的问题。（@yes-a-hit）
• 修复消息文字选择菜单中的分享及第三方翻译等文本处理入口不可用的问题。
• 修复部分页面切换后滚动位置及局部界面状态丢失的问题。
• 修复创建单聊、群聊和故事时，键盘遮挡输入内容及底部操作区域的问题。
• 修复背景图片因采样分辨率不足而显得模糊的问题，改善高分辨率屏幕上的背景清晰度。
• 修复完整备份恢复过程中的数据库兼容性校验问题，避免有效备份导入失败。
• 修复编辑消息时思考链与正文未隔离、修改正文可能影响原有思考链的问题。（@yes-a-hit）

ℹ️ 使用说明与注意事项

• 备份与恢复仍为试验性功能： 跨设备与跨版本恢复的完整可靠性尚未得到充分验证。恢复会完全替换现有应用数据，不会合并，并自动重启应用；操作前请先备份需要保留的数据。
• 备份 ZIP 未加密： 文件包含 API 密钥、自定义请求头、私密对话、附件及已保存的请求日志，请妥善保管，勿公开分享。
• 搜索加速默认关闭，普通搜索仍可使用。开启加速需要额外时间与存储空间；正文搜索不包含思考链和内部摘要消息。
• OpenAI Responses 是否可用取决于服务商与模型是否支持该协议，其最大输出额度包含推理 Token。模型级 System Prompt 不参与上下文预算裁剪，请自行预留额度。
• 对话框后方页面的模糊效果取决于设备与系统支持；不可用时会使用更实的底色或更深的遮罩代替。
• 请求日志需在应用内开启调试模式后采集。查看和导出的内容以已保存日志为准，超出采集上限的内容无法通过导出恢复；分享前请检查其中的私密信息。

---

English Version (英文)

✨ What's New

• Theme Colors and Custom Backgrounds:
  • Added six color palettes: Classic Blue, Sakura Pink, Forest Green, Twilight Purple, Parchment Brown, and Graphite Gray, plus system dynamic colors.
  • Set background images for general pages, single chats, group chats, and stories, with adjustable image position, blur, and background scrim.
  • Adjust transparency and background blur separately for menus, bottom action bars, and content areas. Single chats, group chats, and stories can follow general settings or use independent settings, with previews available.
  • Assign a background to a character card and optionally use it as the preferred background in single chats.
  • Added dialog appearance settings for dialog background transparency and blur, control background transparency, and dimming and blur behind dialogs.
• Content Search and Character Filtering:
  • Added a search entry on the home page. Filter by single chats, group chats, or stories, and search all content, titles and characters, or body text.
  • Search all saved chat messages and story chapters, then jump to matching historical messages or chapters, regardless of which messages or chapter are currently loaded.
  • Filter characters by name, description, or tags when creating single chats, group chats, and stories, while retaining existing selections.
  • Added optional search acceleration. Enable it manually and complete index building to activate it. Index building can be cancelled, and disabling acceleration clears the index and reclaims storage.
• Full Data Backup and Restore: (@onemind66)
  • Added a Backup and Restore page to export application data as a ZIP archive, including characters, single chats, group chats, stories, World Books, Regex scripts, model configurations, preferences, and attachments.
  • Backups include API keys, Token usage, and saved request logs, with data organized into JSON / JSONL files for inspection and storage.
  • Restore data from a backup, with archive integrity and database compatibility checks before replacement and an automatic app restart afterward.
• Model Protocols and Conversation Controls:
  • Added a dedicated native OpenAI Responses protocol, supporting text and image input, streaming replies, reasoning content parsing, and provider-reported usage.
  • New OpenAI templates use Responses by default. Existing configurations retain their protocol, and protocols can also be switched manually in advanced settings.
  • Added a model-level System Prompt that is prepended to requests using that model, including chats, story generation, summaries, and connection tests.
  • Added a per-conversation option to include character card example dialogues in single and group chats. Configure it during creation or in conversation settings; the global example dialogue policy still applies.
• Story Reading and Message Editing:
  • Added themed Markdown previews for story chapters, with a switch back to source editing and shortcuts to the top and bottom. (@onemind66)
  • Improved highlighting and navigation for story search results, making it easier to return to the source and continue editing. (@onemind66)
  • Added a Separate Reasoning setting. By default, editing changes only the message body and preserves existing reasoning blocks; disable it to edit the full message.
• Debug Log Viewing and Export:
  • The JSON viewer now offers tree and source views for inspecting saved requests and responses.
  • Export the viewed content as formatted JSON, or export the saved raw text when formatting is unavailable.
  • Streaming response logs are now organized by event, preserving event metadata and response payloads for troubleshooting.

🎨 Improvements

• Interface and Interaction Consistency:
  • Unified tabs across single chats, group chats, stories, World Books, and Regex editing, and refined layouts for settings, model configurations, backup and restore, and creation pages.
  • Unified World Book selection and management components, creation-page action bars, and button transparency and blur effects, with improved light-theme visuals.
  • Removed some repetitive descriptions, refined the branch-creation icon, and expanded the model configuration management entry's touch target in settings.
• Parameters and Runtime Overhead:
  • Token shortcuts and model preset budgets now use decimal values for easier configuration.
  • When the in-app debug mode is off, request log capture, log content construction, and saving are skipped to reduce overhead.

🐛 Bug Fixes

• Fixed group chat messages not displaying the correct character avatars. (@yes-a-hit)
• Fixed unavailable sharing and third-party text-processing actions, such as translation, in the message text selection menu.
• Fixed scroll positions and local interface state being lost when switching between some pages.
• Fixed the keyboard obscuring input fields and bottom actions when creating single chats, group chats, and stories.
• Fixed background images appearing blurry due to insufficient sampling resolution, improving clarity on high-resolution screens.
• Fixed database compatibility checks during full backup restoration that could cause valid backups to fail to import.
• Fixed message editing not separating reasoning blocks from the body, which could cause body edits to affect existing reasoning content. (@yes-a-hit)

ℹ️ Notes and Precautions

• Backup and restore remains experimental: Full reliability across devices and app versions has not been sufficiently verified. Restoring completely replaces existing application data without merging and automatically restarts the app. Back up any data you want to retain first.
• Backup ZIP files are not encrypted: They contain API keys, custom headers, private conversations, attachments, and saved request logs. Store them securely and do not share them publicly.
• Search acceleration is off by default; ordinary search remains available. Acceleration requires additional time and storage to build its index. Body search excludes reasoning blocks and internal summary messages.
• OpenAI Responses requires support from the selected provider and model, and its maximum output budget includes reasoning tokens. Model-level System Prompts are outside context budget trimming, so reserve enough capacity for them.
• Blur behind dialogs depends on device and system support. A more opaque background or darker scrim is used when it is unavailable.
• Request logs are captured only when the in-app debug mode is enabled. Viewing and exporting use the saved log content; exporting cannot recover content beyond capture limits. Check for private information before sharing.