Keyhide 是一个运行在本地的 AI 中转代理服务。它为你生成一组伪装的 URL 和 API Key，你的开发项目使用这组伪装凭证调用 Keyhide，Keyhide 再转发请求到你真实的上游 AI 平台（OpenAI、Anthropic 或任何兼容的中转站）。
核心价值：当你把项目上传到 GitHub 时，代码里只有伪装凭证，真实的 API Key 永远留在本地，不会泄露。
| 场景 | 说明 |
|---|---|
| 密钥保护 | 开发 AI 应用或使用需要外接 URL 与 API Key 的 AI 工具时，通过 Keyhide 中转调用，真实 Key 永不暴露在项目代码中 |
| 聚合统一调用 | 将所有平台的 URL 和 API Key 聚合到本地统一入口，Keyhide 根据模型名自动路由到对应平台，单平台多 Key 额度用完自动接力切换，全程无需改配置改代码 |
| 开发/生产隔离 | 切换上游平台只需改 Keyhide 配置，无需修改应用代码，开发环境用便宜模型、生产环境用强模型一键切换 |
| 请求日志与监控 | 自动记录每次调用的耗时、Token 用量、成功/失败状态，方便排查问题和统计成本 |
| 功能 | 说明 |
|---|---|
| 密钥保护 | 生成伪装 URL 和 API Key，真实密钥永不暴露给项目代码 |
| 多平台聚合 | 同时接入多个 AI 平台（OpenAI / Anthropic / 自定义），统一入口 |
| 多 Key 轮询 | 单个平台配置多个 Key，支持轮询 / 优先级 / 加权 / 随机策略 |
| 自动故障转移 | Key 报错或超时自动切换到下一个可用 Key，可选自动禁用故障 Key |
| 额度耗尽检测 | 429 状态码自动标记 Key 为已耗尽，跳过使用 |
| 全格式兼容 | 支持 OpenAI 和 Anthropic 全部 API 端点（Chat、Embeddings、Images、Audio、Files、Assistants、Threads 等） |
| 流式响应 | 完整支持 SSE 流式转发，兼容 ChatGPT 流式调用 |
| 加密存储 | 所有配置数据加密存储，与机器绑定，文件被拷走也无法解密 |
| 模型列表 | 自动获取上游模型列表，支持搜索和一键复制 |
| 请求日志 | 记录每次请求的状态、耗时、Token 用量，支持筛选和统计 |
| 数据管理 | 支持重置统计数据和清除所有数据，灵活管理本地存储 |
| 批量 Key 操作 | 一键全部启用 / 禁用某个平台的所有 Key |
- 前往 Releases 页面下载最新版本
- 运行
Keyhide Setup 1.0.0.exe安装，或使用Keyhide-Portable-1.0.0.exe免安装运行 - 打开 Keyhide，进入「上游平台」页面
- 点击「添加平台」，填写：
- 平台名称：例如
OpenAI 官方 - 平台类型：选择
OpenAI 格式或Anthropic 格式 - Base URL：例如
https://api.openai.com（不需要带/v1） - API Key：填入你的真实 Key
- 平台名称：例如
- 点击「获取模型」拉取可用模型列表
- 进入「代理服务」页面，启动代理服务
- 复制伪装 URL 和伪装 API Key，用它们替换你项目中的真实凭证
# 1. 克隆仓库
git clone https://github.com/relliex/Proxy_Keyhide.git
cd Proxy_Keyhide
# 2. 安装依赖
npm install
# 3. 开发模式运行
npm run dev
# 4. 打包构建
npm run build:win # Windows
npm run build:mac # macOS
npm run build:linux # Linux配置好 Keyhide 后，只需把项目中的 API 配置改为伪装凭证：
Python (OpenAI SDK)
from openai import OpenAI
client = OpenAI(
api_key="sk-keyhide-xxxxxxxxxxxx", # Keyhide 生成的伪装 Key
base_url="http://127.0.0.1:7860/v1" # Keyhide 的本地地址
)
response = client.chat.completions.create(
model="gpt-4o",
messages=[{"role": "user", "content": "Hello!"}]
)Python (Anthropic SDK)
from anthropic import Anthropic
client = Anthropic(
api_key="sk-keyhide-xxxxxxxxxxxx",
base_url="http://127.0.0.1:7860"
)
message = client.messages.create(
model="claude-sonnet-4-20250514",
max_tokens=1024,
messages=[{"role": "user", "content": "Hello!"}]
)Node.js
import OpenAI from 'openai'
const client = new OpenAI({
apiKey: 'sk-keyhide-xxxxxxxxxxxx',
baseURL: 'http://127.0.0.1:7860/v1'
})| 模式 | 说明 |
|---|---|
| 自动 (auto) | 优先使用默认平台，若不支持该模型则自动查找其他平台 |
| 按模型 (by-model) | 严格根据模型名查找支持该模型的平台 |
| 按路径 (by-path) | 所有请求都使用默认平台 |
你也可以在请求路径中指定平台：http://127.0.0.1:7860/{platformId}/v1/chat/completions
- 出错自动禁用 Key：开启后，当 Key 请求超时或返回错误时立即禁用该 Key
- 被禁用的 Key 可在「上游平台」页面手动重新启用
所有配置数据加密存储在本地，与机器绑定。
数据与机器绑定，即使文件被复制到其他电脑也无法解密。
- Electron 31 — 跨平台桌面应用框架
- React 18 + TypeScript — 前端 UI
- Vite 5 — 构建工具
- Fastify 4 — HTTP 代理服务器
- Tailwind CSS 3 — 样式
- Zustand — 状态管理
Proxy_Keyhide/
├── electron/
│ ├── main/
│ │ ├── database.ts # 数据存储
│ │ ├── proxy-server.ts # 代理服务器（Fastify）
│ │ ├── ipc.ts # IPC 通信
│ │ ├── window.ts # 窗口管理
│ │ ├── index.ts # 主进程入口
│ │ └── upstream/
│ │ └── manager.ts # 负载均衡与故障转移
│ ├── preload/
│ │ └── index.ts # 预加载脚本
│ └── shared/
│ └── types.ts # 共享类型定义
├── src/
│ ├── components/ # UI 组件
│ ├── pages/ # 页面（Dashboard/Platforms/Proxy/Logs/Settings）
│ ├── stores/ # Zustand 状态管理
│ └── main.tsx # 渲染进程入口
├── resources/
│ └── icon.png # 应用图标
└── package.json
- 默认绑定
127.0.0.1，仅本机访问，不暴露到公网 - 所有数据加密存储，与机器绑定，文件被拷走也无法解密
- 伪装 API Key 仅存在于本地加密文件中
- 上游真实 Key 不会通过代理响应泄露给调用方
⚠️ 如果将监听地址改为0.0.0.0以共享给局域网，请注意 API Key 会以 HTTP 明文传输，建议配合防火墙使用。
Keyhide is a local AI relay proxy service. It generates a pair of masked URL and API Key for your development projects. Your projects use these masked credentials to call Keyhide, which then forwards requests to your real upstream AI platforms (OpenAI, Anthropic, or any compatible relay service).
Core value: When you upload your project to GitHub, the code only contains masked credentials — your real API keys never leave your local machine.
| Use Case | Description |
|---|---|
| Key Protection | When developing AI apps or using AI tools that require external URL and API key, relay through Keyhide to keep your real keys out of project code |
| Unified Aggregation | Aggregate all platform URLs and API keys into a single local entry point. Keyhide auto-routes requests to the correct platform by model name, and auto-relays across multiple keys when quota runs out — no config or code changes needed |
| Dev/Prod Isolation | Switch upstream platforms by simply changing Keyhide config — no code changes needed. Use cheap models in dev, powerful models in prod with one click |
| Request Logging & Monitoring | Automatically records duration, token usage, and success/failure status for each call — easy debugging and cost tracking |
| Feature | Description |
|---|---|
| Key Protection | Generates masked URL and API Key; real keys are never exposed to project code |
| Multi-Platform | Connect multiple AI platforms (OpenAI / Anthropic / custom) with a unified entry point |
| Multi-Key Rotation | Configure multiple keys per platform with round-robin / priority / weighted / random strategies |
| Auto Failover | Automatically switches to the next available key on errors or timeouts; optionally auto-disables faulty keys |
| Quota Detection | Automatically marks keys as exhausted on 429 status and skips them |
| Full API Compatibility | Supports all OpenAI and Anthropic API endpoints (Chat, Embeddings, Images, Audio, Files, Assistants, Threads, etc.) |
| Streaming Support | Full SSE streaming passthrough, compatible with ChatGPT-style streaming calls |
| Encrypted Storage | All configuration data encrypted and machine-bound — files cannot be decrypted even if copied |
| Model List | Auto-fetches upstream model lists with search and one-click copy |
| Request Logging | Records status, duration, and token usage for each request with filtering and statistics |
| Data Management | Reset statistics or clear all data with confirmation dialogs |
| Bulk Key Operations | Enable / disable all keys for a platform with one click |
- Go to the Releases page and download the latest version
- Run
Keyhide Setup 1.0.0.exeto install, or useKeyhide-Portable-1.0.0.exefor portable mode - Open Keyhide and navigate to the Upstream Platforms page
- Click Add Platform and fill in:
- Platform Name: e.g.,
OpenAI Official - Platform Type: Select
OpenAI FormatorAnthropic Format - Base URL: e.g.,
https://api.openai.com(no/v1suffix needed) - API Key: Enter your real key
- Platform Name: e.g.,
- Click Fetch Models to retrieve available models
- Go to the Proxy Service page and start the proxy
- Copy the masked URL and masked API Key — use them to replace real credentials in your project
# 1. Clone the repository
git clone https://github.com/relliex/Proxy_Keyhide.git
cd Proxy_Keyhide
# 2. Install dependencies
npm install
# 3. Run in development mode
npm run dev
# 4. Build and package
npm run build:win # Windows
npm run build:mac # macOS
npm run build:linux # LinuxAfter configuring Keyhide, simply replace your project's API credentials with the masked ones:
Python (OpenAI SDK)
from openai import OpenAI
client = OpenAI(
api_key="sk-keyhide-xxxxxxxxxxxx", # Masked key from Keyhide
base_url="http://127.0.0.1:7860/v1" # Keyhide local address
)
response = client.chat.completions.create(
model="gpt-4o",
messages=[{"role": "user", "content": "Hello!"}]
)Python (Anthropic SDK)
from anthropic import Anthropic
client = Anthropic(
api_key="sk-keyhide-xxxxxxxxxxxx",
base_url="http://127.0.0.1:7860"
)
message = client.messages.create(
model="claude-sonnet-4-20250514",
max_tokens=1024,
messages=[{"role": "user", "content": "Hello!"}]
)Node.js
import OpenAI from 'openai'
const client = new OpenAI({
apiKey: 'sk-keyhide-xxxxxxxxxxxx',
baseURL: 'http://127.0.0.1:7860/v1'
})| Mode | Description |
|---|---|
| Auto | Uses default platform first; auto-finds other platforms if the model is unsupported |
| By Model | Strictly finds platforms that support the requested model |
| By Path | All requests use the default platform |
You can also specify a platform in the request path: http://127.0.0.1:7860/{platformId}/v1/chat/completions
- Auto-disable Key on Error: When enabled, keys are immediately disabled on timeout or error responses
- Disabled keys can be manually re-enabled in the Upstream Platforms page
All configuration data is encrypted and stored locally, machine-bound.
Data is machine-bound — even if files are copied to another computer, they cannot be decrypted.
- Electron 31 — Cross-platform desktop framework
- React 18 + TypeScript — Frontend UI
- Vite 5 — Build tool
- Fastify 4 — HTTP proxy server
- Tailwind CSS 3 — Styling
- Zustand — State management
Proxy_Keyhide/
├── electron/
│ ├── main/
│ │ ├── database.ts # Data storage
│ │ ├── proxy-server.ts # Proxy server (Fastify)
│ │ ├── ipc.ts # IPC communication
│ │ ├── window.ts # Window management
│ │ ├── index.ts # Main process entry
│ │ └── upstream/
│ │ └── manager.ts # Load balancing & failover
│ ├── preload/
│ │ └── index.ts # Preload script
│ └── shared/
│ └── types.ts # Shared type definitions
├── src/
│ ├── components/ # UI components
│ ├── pages/ # Pages (Dashboard/Platforms/Proxy/Logs/Settings)
│ ├── stores/ # Zustand state management
│ └── main.tsx # Renderer process entry
├── resources/
│ └── icon.png # App icon
└── package.json
- Binds to
127.0.0.1by default — localhost only, not exposed to the public internet - All data encrypted and machine-bound — files cannot be decrypted even if copied
- Masked API Key exists only in the local encrypted file
- Upstream real keys are never leaked through proxy responses
⚠️ If you change the listen address to0.0.0.0to share with your LAN, note that API keys will be transmitted in HTTP plaintext. Use with firewall protection.
Made with care for developers who value security.