RPClient

Releases

• Sep 18, 2026 2026.3.1
  # 📦 RPClient 2026.3.1 ### 中文版本 (Chinese) #### ✨ 新增功能 - **单聊与群聊图片消息：** - 支持发送图片、图文混合消息及纯图片消息，每条新消息最多可添加四张图片。 - 新增图片查看器，支持缩放、双击、左右切换、缩略图导航与保存图片。 - 提供自动、原图和自定义图片发送策略，可按需设置图片大小及宽高上限。

  More…

  - 支持通过 OpenAI-Compatible、Anthropic Messages 与 Gemini 协议发送图片，模型配置中可设置图片输入能力与 Token 估算策略。 - 图片消耗纳入上下文与摘要预算，Prompt 检查器可查看图片相关信息及省略原因。 - 用户消息编辑支持添加、删除和排序图片；导入的角色及旁白图片可保留、删除和排序，暂不开放添加入口。 - **聊天图片导入与导出：** - 单聊 JSONL 归档现可携带图片，导出后重新导入可恢复消息附件。 - 支持导入 SillyTavern 的图片消息；遇到外部图片引用时，可选择本地资源目录匹配图片，缺失项会统计并跳过。 - 图片归档支持压缩和重复资源去重，减少重复图片占用的文件空间。 - 支持保留导入的角色及旁白图片，并在创建分支时一同复制。 - **Token 用量统计：** - 新增用量统计页面，展示输入、输出、总 Token 数和成功请求次数。 - 支持按时间范围筛选、查看柱状图与趋势图，以及按模型和 API 主机汇总用量。 - 最近请求明细可展开查看耗时、计数来源及可用的缓存和推理用量信息，支持清空统计记录。 - 模型配置新增服务端用量开关及本地 Token 预估器选择；新建 DeepSeek 配置默认启用服务端用量。 - **个性化与配置：** - 新增跟随系统、浅色、深色三种主题模式。 - 模型配置支持克隆，方便在保留已有参数的基础上创建配置副本。 - 新增葡萄牙语（巴西）、意大利语、越南语和印度尼西亚语界面，并补齐已有语言的新功能文案。 #### 🎨 体验优化 - **长对话与上下文：** - 单聊、群聊的历史消息改为分页加载，并优化上下文构建与 Token 计算，减少长会话中的重复处理。 - 新增参与模型请求的历史消息数量设置，默认使用最近 500 条，设为 0 可取消此限制；本地聊天记录不会因此删除。 - **聊天与编辑体验：** - 聊天正文支持长按选择和复制文字。 - 新增列表侧边滚动指示器，支持拖动定位和自动隐藏。 - 优化聊天输入区、键盘展开与横屏布局，统一世界书快捷管理对话框，并调整世界书条目、模型配置和故事编辑页面。 - **提示与兼容性：** - 改善模型未配置、请求失败及摘要额度不足等情况的提示和设置引导。 - 优化摘要生成失败、空结果和截断结果的处理，避免不完整内容覆盖有效摘要。 - @Equaraphanus 支持用户安装的系统级 CA 证书，方便连接使用自定义证书链的服务。 #### 🐛 问题修复 - 修复部分横屏页面被系统导航栏遮挡的问题。 - 修复故事导出时，世界书条目过多导致数据库查询失败的问题。 --- ### English Version (英文) #### ✨ What's New - **Images in Single and Group Chats:** - Send images with text or as image-only messages, with up to four images per new message. - Added an image viewer with zoom, double-tap gestures, swipe navigation, thumbnails, and image saving. - Choose automatic, original, or custom image processing, with configurable file size and dimension limits. - Send images through OpenAI-compatible, Anthropic Messages, and Gemini protocols, with configurable image input capabilities and token estimation strategies. - Image costs now participate in context and summary budgets. The Prompt Inspector shows image information and reasons for omission. - Add, remove, and reorder images when editing user messages. Existing character and narrator images can be retained, removed, or reordered, while the add-image action remains unavailable for those messages. - **Chat Image Import and Export:** - Single-chat JSONL archives can now include images and restore attachments when imported back into RPClient. - Import SillyTavern image messages and select a local resource folder to resolve external image references. Missing images are counted and skipped. - Archives support compression and resource deduplication to reduce space used by repeated images. - Imported character and narrator images are preserved when creating chat branches. - **Token Usage Statistics:** - Added a usage dashboard showing input, output, total tokens, and successful request counts. - Filter by time period, switch between bar and trend charts, and view totals grouped by model and API host. - Expand recent requests to inspect duration, counting sources, and available cache and reasoning usage. Usage records can also be cleared. - Added a provider-reported usage setting and local tokenizer selection. New DeepSeek configurations enable provider-reported usage by default. - **Personalization and Configuration:** - Choose between system, light, and dark themes. - Clone model configurations to create a copy while retaining existing parameters. - Added Brazilian Portuguese, Italian, Vietnamese, and Indonesian translations, and updated existing languages for the new features. #### 🎨 Improvements - **Long Conversations and Context:** - Added paginated history loading and improved context construction and token calculation in single and group chats to reduce repeated processing. - Added a prompt history limit, using the latest 500 messages by default. Set it to 0 to disable the limit; stored chat history is not deleted. - **Chat and Editing Experience:** - Long-press message text to select and copy it. - Added side scroll indicators with drag navigation and automatic hiding. - Refined chat input, keyboard, and landscape layouts; unified quick World Book management dialogs and improved entry, model configuration, and story editing screens. - **Feedback and Compatibility:** - Improved guidance for missing model configurations, failed requests, and insufficient summary budgets. - Improved handling of failed, empty, and truncated summary responses to avoid replacing valid summaries with incomplete content. - @Equaraphanus Added support for user-installed system CA certificates when connecting to services with custom certificate chains. #### 🐛 Bug Fixes - Fixed system navigation bars obscuring some landscape layouts. - Fixed story export failing when too many World Book entries were included in a database query.
• Aug 27, 2026 2026.2.3
  版本迭代：2026.2.3 # 📦 RPClient 2026.2.3 ### 中文版本 (Chinese) #### ✨ 新增功能 - 故事创作思考过程展示： - 故事续写新增实时生成状态卡，分阶段展示上下文准备、等待响应、模型连接、构思、正文写入与保存状态。 - 支持查看模型返回的思考内容，并可展开或收起详情；思考内容与章节正文分离，不会混入最终稿件。

  More…

  - 兼容 OpenAI-Compatible、Anthropic Messages 与 Gemini 协议返回的推理内容。 - 新增生成耗时统计、慢响应提示及流式思考内容自动滚动。 - 角色卡批量导入： - 文件选择器现可一次选择并导入多张角色卡。 - 提供解析与保存阶段的实时进度，并在完成后汇总成功及失败数量。 - 单个文件导入失败不会中断整个批次；内嵌世界书低预算提示也已适配批量处理。 #### 🎨 体验优化 - 模型配置引导： - 单聊、群聊与故事创作页面会明确提示尚未配置模型服务。 - 在发送消息或生成内容前提供配置引导，并可一键跳转至模型配置页面，同时保留当前草稿。 - 默认思考强度调整： - Gemini、Claude、DeepSeek、Grok 与 OpenRouter 的内置高级参数预设默认改为 `low`，降低响应延迟与推理消耗。 - Gemini 预设启用思考摘要返回，以支持故事创作中的思考过程展示。 - 关于页面焕新： - 重新设计页面布局，新增 Google Play 评分、GitHub Issues、开源仓库、开发者联系方式与 MIT License 信息入口。 - 外部应用不可用时可自动回退至网页或复制链接。 - 群聊操作调整： - 将删除群聊操作从顶部工具栏移至设置页面，降低误触风险。 - 多语言完善： - 新增完整的西班牙语界面资源，并补齐其他语言中遗漏的文案。 #### 🐛 问题修复 - 修复沉浸式界面下，Android 三键导航栏或软键盘遮挡输入框、底栏及列表末尾的问题。 - 修复未配置模型服务时，单聊、群聊和故事页面仍错误显示为可用或就绪状态的问题。 --- ### English Version (英文) #### ✨ What's New - Story Generation Reasoning Display: - Added a real-time generation status card that tracks context preparation, response waiting, connection, reasoning, writing, and result application. - Model reasoning can now be expanded or collapsed while remaining completely separate from the final chapter content. - Supports reasoning output from OpenAI-compatible, Anthropic Messages, and Gemini protocols. - Added elapsed-time tracking, slow-response hints, and automatic scrolling for streamed reasoning. - Batch Character Card Import: - Multiple character card files can now be selected and imported at once. - Displays real-time parsing and saving progress, followed by a summary of successful and failed imports. - A failed file no longer interrupts the remaining batch, and embedded lorebook budget confirmation now supports multiple cards. #### 🎨 Improvements - Model Configuration Guidance: - Chat, group chat, and Story Studio now clearly indicate when no model service is configured. - Attempts to send or generate content are intercepted with a setup guide and a shortcut to model settings, while preserving the current draft. - Lower Default Reasoning Effort: - Built-in advanced presets for Gemini, Claude, DeepSeek, Grok, and OpenRouter now default to `low` reasoning effort to reduce latency and inference cost. - Gemini presets now request thought summaries for Story Studio reasoning display. - Redesigned About Page: - Added dedicated entries for Google Play ratings, GitHub Issues, the source repository, developer contact information, and the MIT License. - External links gracefully fall back to a browser or clipboard when the preferred application is unavailable. - Group Chat Actions: - Moved the delete action from the top toolbar into group chat settings to reduce accidental deletion. - Localization: - Added complete Spanish localization and filled in missing translations across existing language packs. #### 🐛 Bug Fixes - Fixed input bars, bottom actions, and list content being obscured by Android three-button navigation or the on-screen keyboard. - Fixed chat, group chat, and Story Studio incorrectly appearing ready when no model provider was configured.
• Aug 27, 2026 2026.2.2
  ## 📦 RPClient 2026.2.2 ### 中文版本 (Chinese) #### ✨ 新增功能 * **全新小说创作系统（Story Studio）**： * 支持「作品 - 分卷 - 章节」三层树状结构管理，支持章节跨卷拖拽排序与移动。 * 支持章节续写引导词持久化保存，并与世界书及登场角色人设联动组装上下文。 * 支持分层式 V2 小说归档格式（`.rpstory`）导入与导出。 * **酒馆（SillyTavern）兼容对话导入导出**：

  More…

  * 支持将单聊与群聊记录无损导出为酒馆标准 `.jsonl` 格式；支持从酒馆导入历史会话 (#3)。 * 导入时提供角色指纹与名称智能匹配绑定；采用流式分页读写，杜绝超长对话导出时的 OOM 崩溃 (#3)。 * **模型独立绑定与分流**： * 支持为特定角色及长文本摘要（Summary）分别绑定独立的模型配置，实现日常对话与摘要记忆的模型解耦与成本优化 (#4)。 * **推理等级与深度思考调节**： * 支持为推理大模型（o1/o3、Claude 3.7+、Gemini 3.1+ 等）设置推理等级（Reasoning Effort）。 * **OpenRouter 路由与通用请求补丁**： * 支持配置 OpenRouter 供应商路由偏好与通用 Request Body Patch，可向请求体自由追加任意自定义 JSON 参数。 * **高级参数预设与 JSON 高亮**： * 内置常见供应商高级参数预设模板，高级 JSON 配置编辑器新增实时语法高亮。 * **世界书互斥包含组**： * 支持为条目配置互斥包含组（Inclusion/Exclusion Groups），同类设定单次仅激活最佳词条，避免设定冲突并节省 Token。 * **内置正方形头像裁剪**： * 上传头像时支持直观的正方形裁剪框，自适应相机与图片的 EXIF 旋转方向。 * **单色主题图标**： * 采用矢量资源适配 Android 13+ Material You 单色动态主题图标 (#6)。 #### 🎨 体验优化 * **全局视觉系统重构**：全面重构主题色盘、排版字阶与卡片层次，提升主页、单聊、群聊等核心页面的现代感。 * **标准化对话框套件**：重构并统一了确认框、输入框、Prompt 编辑框及代码框等通用弹窗的交互规范与键盘避让。 * **会话与列表排序**：故事与群聊列表支持长按拖拽排序；主页会话列表新增快捷置顶、导出、删除等操作。 * **Prompt 宏高亮**：人设与提示词编辑支持 `{{char}}`、`{{user}}` 等宏语法高亮与快捷插入栏。 * **请求日志加速**：优化长日志虚拟化渲染结构，大幅提升日志列表与预览的滑动流畅度。 * **自动化与应用商店元数据**：规范化 Fastlane 发布元数据与多语言商店截图配置 (#7)。 #### 🐛 问题修复 * 修复世界书关键词在 CJK（中文/日文/韩文）语境下因缺乏词边界导致全字匹配无法触发的问题。 * 修复 Claude 开启推理模式时因附带非默认 `temperature` / `top_p` 参数被上游拒绝的问题。 * 针对官方 ChatGPT 请求适配 `max_completion_tokens`，同时兼容第三方网关的 `max_tokens`。 * 自动清洗 Grok 推理模型不支持的 `stop` 字段，适配 Gemini 思考等级映射差异。 * 修复角色编辑页标签与搜索框在部分窄屏和多语言环境下的异常折行问题。 * 修复会话导入时的异步状态竞争与存储授权残留。 --- ### English Version (英文) #### ✨ What's New * **Brand New Story Studio**: * Full hierarchical structure support (`Novel - Volume - Chapter`), with cross-volume drag-and-drop reordering. * Persistent continuation guidance with real-time lorebook keyword activation and character persona assembly. * Support for the structured V2 story archive format (`.rpstory`) for seamless backup and sharing. * **SillyTavern-Compatible Chat Import & Export**: * Full bidirectional `.jsonl` chat import/export compatible with SillyTavern formats (#3). * Smart character matching based on name and fingerprint; stream-based paginated export to completely prevent OOM crashes on large chats (#3). * **Per-Character & Summary Model Binding**: * Allows assigning independent model configurations to specific characters as well as summary generation (#4). * **Reasoning Effort Control**: * Adjust reasoning effort / thinking levels for reasoning-capable models (o1, o3, Claude 3.7+, Gemini 3.1+, etc.). * **OpenRouter Routing & Request Body Patching**: * Configurable OpenRouter provider routing preferences and dynamic Request Body Patching via custom JSON. * **Advanced Presets & JSON Syntax Highlighting**: * Pre-configured vendor parameter presets and a real-time syntax-highlighted JSON code editor. * **Lorebook Inclusion / Exclusion Groups**: * Added mutual exclusion group support to eliminate conflicting lorebook activations and conserve context tokens. * **Built-in Square Avatar Cropping**: * Intuitive square cropping tool when selecting avatars, with automatic EXIF orientation handling. * **Monochrome Themed Icon**: * Migrated to vector drawables to natively support Android 13+ Material You monochrome themed icons (#6). #### 🎨 Improvements * **Global Visual Redesign**: Complete overhaul of the color palette, typography hierarchy, and card elevation across Home, Chat, and Story interfaces. * **Standardized Dialog Suite**: Unified look, feel, and keyboard-avoidance behavior for all dialogs (Confirm, Input, Prompt Editor, Code Editor, Action Lists). * **List Reordering & Quick Actions**: Long-press drag-and-drop sorting for stories and group chats; quick session actions on the home screen. * **Prompt Macro Highlighting**: Syntax highlighting for placeholder macros (`{{char}}`, `{{user}}`) with a dedicated quick-insert action bar. * **Request Log Performance**: Lightweight overview rendering for butter-smooth scrolling through large LLM request logs. * **Fastlane & Store Metadata**: Refined metadata directory structure and localized store screenshots (#7). #### 🐛 Bug Fixes * Fixed lorebook keyword activation failure in CJK (Chinese, Japanese, Korean) text under whole-word matching. * Fixed upstream HTTP 400 errors with Claude reasoning models by automatically filtering non-default `temperature` and `top_p`. * Seamlessly adapted to `max_completion_tokens` for official ChatGPT models while preserving `max_tokens` for third-party gateways. * Filtered unsupported `stop` tokens for Grok models and aligned Gemini thinking level mappings. * Fixed text wrapping issues in character tags and search bars on narrow screens and localized layouts. * Resolved asynchronous state race conditions and persistent URI permission leaks during chat imports.