iOS app · communication assistant iOS 应用 · 沟通辅助
Social Copilot Social Copilot
An iOS app that uses a specific contact, relationship, and message to generate three reply suggestions without reading or sending messages for the user.
一个面向具体联系人和具体消息的 iOS 沟通辅助工具：结合关系和个人偏好生成三条回复建议，但不读取、也不代发消息。
产品
针对一个联系人和一条消息
Social Copilot 不是聊天软件。用户主动输入或粘贴一条消息，选择联系人、关系和沟通目标，App 再给出三种可以直接复制的回复建议。
它不会读取微信或短信，也不会自动发送。AI 只负责给建议，最后怎么说仍然由用户决定。
输入
文字、语音和聊天截图
消息可以键入、粘贴、语音转写，也可以从聊天截图中通过本机 OCR 提取。联系人页面保存关系类型、备注和用户手动维护的记忆，个人表达偏好同样可以查看和修改。
生成时会把这些明确的上下文组合起来，而不是在后台自动推断用户的人际关系。
实现
SwiftUI 客户端和一个受限的模型代理
iOS 客户端使用 SwiftUI、Vision OCR、Speech 和 AVFoundation。本地数据以受保护的原子 JSON snapshot 保存，并包含旧数据迁移和失败恢复。
模型请求经过 Node.js 代理，API key 留在服务端。代理包含请求大小、超时和频率限制，也不会记录 prompt 或请求正文。项目同时有 iOS 单元测试、UI 测试和后端测试。
状态
完整 MVP，正在准备上架
核心流程已经可以运行：选择联系人、输入消息、生成三条建议、复制回复，并按需保存历史。当前工作主要是实机视觉和权限检查，以及补齐隐私政策、支持页面等上架材料。
THE PRODUCT
One contact and one message at a time
Social Copilot is not a messaging app. The user enters or pastes a message, chooses the contact, relationship, and communication goal, and receives three replies that can be copied into another app.
It does not read WeChat or SMS and never sends a message automatically. The AI suggests; the user decides what to say.
INPUT
Text, speech, and chat screenshots
A message can be typed, pasted, dictated, or extracted from a screenshot with on-device OCR. Contact pages hold relationship details, notes, and memories maintained by the user; writing preferences are also visible and editable.
Generation combines that explicit context instead of silently inferring the user's relationships in the background.
THE BUILD
A SwiftUI client and a constrained model proxy
The iOS client uses SwiftUI, Vision OCR, Speech, and AVFoundation. Local data is stored in protected atomic JSON snapshots with migration and failure recovery.
Model requests go through a Node.js proxy so the API key remains server-side. The proxy enforces request-size, timeout, and rate limits without logging prompts or request bodies. The project includes iOS unit and UI tests as well as backend tests.
STATUS
A working MVP, now preparing for release
The main flow is complete: choose a contact, enter a message, generate three suggestions, copy a reply, and optionally keep the history. The remaining work is device-level visual and permission QA plus the privacy and support pages required for release.