diff --git a/AGENTS.md b/AGENTS.md
index 34c2738..8ba2d0a 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -42,6 +42,30 @@ pnpm run dev
 - **`ASSETS` binding name must be `ASSETS`** (matched in `src/index.ts` as `env.ASSETS`).
 - **Base64 MIME parts** are decoded with `atob()` in `extractMailBodies()`. Cloudflare Workers runtime supports `atob`/`btoa`.
 
+## AI Provider Configuration
+
+The AI layer uses a single unified function `callProvider(config: ProviderConfig, prompt)` in `src/index.ts`. All provider-specific logic is contained there — do not add new per-provider methods.
+
+**Env vars (set in `wrangler.toml` `[vars]` or as Cloudflare Secrets):**
+
+| Variable | Required | Description |
+|---|---|---|
+| `AI_BASE_URL` | ✅ | Provider base URL, no trailing slash |
+| `AI_API_KEY` | ✅ | API key (use Secret in production) |
+| `AI_API_FORMAT` | ✅ | `openai` \| `responses` \| `anthropic` |
+| `AI_MODEL` | ✅ | Model ID |
+| `AI_FALLBACK_BASE_URL` | optional | Fallback provider base URL |
+| `AI_FALLBACK_API_KEY` | optional | Fallback API key |
+| `AI_FALLBACK_API_FORMAT` | optional | Fallback format |
+| `AI_FALLBACK_MODEL` | optional | Fallback model ID |
+
+Fallback is only active when all four `AI_FALLBACK_*` vars are set. Primary retries 3× before fallback is attempted (1×).
+
+**Format → endpoint mapping:**
+- `openai` → `POST /v1/chat/completions` (OpenAI, Gemini OpenAI-compat, DeepSeek, Groq, …)
+- `responses` → `POST /v1/responses` (OpenAI Responses API)
+- `anthropic` → `POST /v1/messages` (Anthropic Claude direct; sends `anthropic-version: 2023-06-01`, `max_tokens: 1024`)
+
 ## Coding Style
 
 - Tabs, LF, UTF-8, no trailing whitespace (see `.editorconfig`).
diff --git a/CLAUDE.md b/CLAUDE.md
index df794d2..902a8b4 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -23,6 +23,12 @@ pnpm run test
 3. `src/index.html` is a live fallback. Do not delete.
 4. DOMPurify + sandboxed iframe on all email HTML rendering. Do not relax.
 5. Basic Auth gate is first in `WorkerEntrypoint.fetch()`. Do not move or bypass.
+6. All AI calls go through `callProvider()`. Do not add per-provider methods alongside it.
+
+## AI Provider Env Vars
+
+Required: `AI_BASE_URL`, `AI_API_KEY`, `AI_API_FORMAT` (`openai`|`responses`|`anthropic`), `AI_MODEL`.
+Optional fallback: `AI_FALLBACK_BASE_URL`, `AI_FALLBACK_API_KEY`, `AI_FALLBACK_API_FORMAT`, `AI_FALLBACK_MODEL` (all four or none).
 
 ## Frontend Component Rules
 
diff --git a/README.md b/README.md
index 5b5810a..f20732e 100644
--- a/README.md
+++ b/README.md
@@ -2,10 +2,9 @@
 
 [English](https://github.com/TooonyChen/AuthInbox/blob/main/README.md) | [简体中文](https://github.com/TooonyChen/AuthInbox/blob/main/README_CN.md)
 
-**Auth Inbox** is an open-source project that securely receive and views authentication emails using [Cloudflare](https://cloudflare.com/)'s free serverless services. It automatically processes incoming emails, extracts verification codes or links, and stores them in a database. A user-friendly web interface is provided for administrators to easily review the extracted information. AuthInbox also supports real-time notifications via Bark, making it a comprehensive and hassle-free solution for email authentication management.
-
-Don't wanna receive ads and spams on your main email? Want a bunch of alternative email for register services and websites? Try this **secure**, **serverless**, **light** service!
+**Auth Inbox** is an open-source, self-hosted email verification code platform built on [Cloudflare](https://cloudflare.com/)'s free serverless services. It automatically processes incoming emails, filters out promotional mail before hitting the AI, extracts verification codes or links, and stores them in a database. A modern React dashboard lets administrators review extracted codes, inspect raw emails, and render HTML email previews — all protected by Basic Auth.
 
+Don't want ads and spam in your main inbox? Need a bunch of alternative addresses for signups? Try this **secure**, **serverless**, **lightweight** service!
 
 [![Deploy to Cloudflare Workers](https://deploy.workers.cloudflare.com/button)](https://deploy.workers.cloudflare.com/?url=https://github.com/TooonyChen/AuthInbox)
 
@@ -15,205 +14,177 @@ Don't wanna receive ads and spams on your main email? Want a bunch of alternativ
 
 ## Table of Contents 📑
 
-- [Features](#features)
-- [Technologies Used](#technologies-used)
-- [Installation](#installation)
-- [License](#license)
-- [Screenshots](#Screenshots)
+- [Features](#features-)
+- [Technologies Used](#technologies-used-)
+- [Installation](#installation-)
+- [License](#license-)
+- [Screenshots](#screenshots-)
 
 ---
 
 ## Features ✨
 
-- **Email Processing**: Automatically captures and stores incoming emails.
-- **Code Extraction**: Utilizes AI to extract verification codes, links, and organization names from emails.
-- **Secure Front-End**: Provides a web interface protected by Basic Access Authentication for viewing extracted codes.
-- **Real-Time Notifications**: Optionally sends notifications via Bark when new codes are extracted.
-- **Database Integration**: Stores raw and processed email data in a Cloudflare D1 Database.
+- **Promotional Filter**: Detects and skips bulk/marketing emails via headers (`List-Unsubscribe`, `Precedence: bulk`, etc.) before calling the AI — saves tokens.
+- **AI Code Extraction**: Uses Google Gemini (with OpenAI as fallback) to extract verification codes, links, and organization names.
+- **Modern Dashboard**: React 18 + shadcn/ui interface with mail list, detail panel, and three tabs — Extracted, Raw Email, Rendered HTML preview.
+- **Safe HTML Preview**: Email HTML is sanitized with DOMPurify and rendered in a sandboxed iframe.
+- **One-click Copy**: Verification codes and links have copy buttons with toast confirmation.
+- **Real-Time Notifications**: Optionally sends Bark push notifications when new codes arrive.
+- **Database**: Stores all raw emails and AI-extracted results in Cloudflare D1.
 
 ---
 
 ## Technologies Used 🛠️
 
-- **Cloudflare Workers**: Serverless platform for handling email processing and web requests.
-- **Cloudflare D1**: Cloudflare's serverless SQL database for storing email data.
-- **TypeScript**: Strongly typed programming language for robust and maintainable code.
-- **Google AI Studio API**: Utilized for extracting relevant information from emails with optimized AI prompts to enhance data accuracy and reliability.
-- **AI Prompt Optimization**: Custom-crafted prompts ensure precise extraction of titles, codes, and topics from varied email formats.
-- **Bark API**: Optional integration for sending real-time notifications.
-- **HTML/CSS**: Front-end interface with responsive and modern design.
-
----
-
-## AI Prompt Optimization 🧠
-
-To ensure accurate extraction of information from incoming emails, we've implemented AI prompt optimization using the Google AI Studio API. By crafting precise and context-aware prompts, the AI can reliably identify and extract key elements such as:
-
-- **Organization Name (Title)**: Identifies the sender's organization or company.
-- **Verification Code/Link**: Extracts codes, links, or passwords necessary for account verification.
-- **Email Topic**: Summarizes the main purpose of the email, such as 'account verification' or 'password reset'.
-
-**Prompt:**
-```plaintext
-Email content: [Insert raw email content here].
-
-Please read the email and extract the following information:
-1. Code/Link/Password from the email (if available).
-2. Organization name (title) from which the email is sent.
-3. A brief summary of the email's topic (e.g., 'account verification').
-
-Format the output as JSON with this structure:
-{
-  "title": "The organization or company that sent the verification code (e.g., 'Netflix')",
-  "code": "The extracted verification code, link, or password (e.g., '123456' or 'https://example.com/verify?code=123456')",
-  "topic": "A brief summary of the email's topic (e.g., 'account verification')",
-  "codeExist": 1
-}
-
-If both a code and a link are present, include both in the 'code' field like this:
-"code": "code, link"
-
-If there is no code, clickable link, or this is an advertisement email, return:
-{
-  "codeExist": 0
-}
-```
+- **Cloudflare Workers** — Serverless runtime for email handling and API.
+- **Cloudflare D1** — SQLite-compatible serverless database.
+- **Cloudflare Email Routing** — Routes incoming emails to the Worker.
+- **React 18 + Vite + Tailwind CSS + shadcn/ui** — Frontend dashboard.
+- **Any OpenAI-compatible / Anthropic AI provider** — Configurable via env vars; Gemini, OpenAI, DeepSeek, Groq, Anthropic, and more.
+- **Bark API** — Optional iOS push notifications.
+- **TypeScript** — End-to-end type safety.
 
 ---
 
 ## Installation ⚙️
-0. **Prerequisites**
 
-	- Create a [Google AI Studio API](https://aistudio.google.com/)
+### Prerequisites
 
-	- Bind a domain to your [Cloudflare](https://dash.cloudflare.com/) account
+- A [Google AI Studio API key](https://aistudio.google.com/)
+- A domain bound to your [Cloudflare](https://dash.cloudflare.com/) account
+- Cloudflare Account ID and API Token from [here](https://dash.cloudflare.com/profile/api-tokens)
+- *(Optional)* [Bark App](https://bark.day.app/) for iOS push notifications
 
-	  - Get Your Cloudflare Account ID from [here](https://dash.cloudflare.com/profile)
+---
 
-      - Generate a Cloudflare Workers API Token from [here](https://dash.cloudflare.com/profile/api-tokens)
+### Option 1 — Deploy via GitHub Actions
 
-	- (Optional) Download the [Bark App](https://bark.day.app/) and get a Bark Token from the App
+1. **Create D1 Database**
 
-1. **Install using Github Pages**
+   Go to [Cloudflare Dashboard](https://dash.cloudflare.com/) → `Workers & Pages` → `D1 SQL Database` → `Create`. Name it `inbox-d1`.
 
-   1. **Creating D1 Database**
+   Open the database → `Console`, paste and run the contents of [`db/schema.sql`](https://github.com/TooonyChen/AuthInbox/blob/main/db/schema.sql).
 
-      1. Go to [Cloudflare Dashboard](https://dash.cloudflare.com/) -> `Workers & Pages` -> `D1 SQL Database` -> `Create`
+   Copy the `database_id` for the next step.
 
-      2. Input Name `inbox-d1` and click `Create`
+2. **Fork & Deploy**
 
-      3. After Creating `inbox-d1`, click in to it and find `Console`
+   [![Deploy to Cloudflare Workers](https://deploy.workers.cloudflare.com/button)](https://deploy.workers.cloudflare.com/?url=https://github.com/TooonyChen/AuthInbox)
 
-      4. Execute the following SQL command from [db/schema.sql](https://github.com/TooonyChen/AuthInbox/blob/main/db/schema.sql) in the console, just copy, paste and execute it.
+   In your forked repository, go to `Settings` → `Secrets and variables` → `Actions` and add:
+   - `CLOUDFLARE_ACCOUNT_ID`
+   - `CLOUDFLARE_API_TOKEN`
+   - `TOML` — use the [comment-free template](https://github.com/TooonyChen/AuthInbox/blob/main/wrangler.toml.example.clear) to avoid parse errors.
 
-      5. Copy the `database_id` and `database_name` for the next step when you configure the `TOML` file
+   Then go to `Actions` → `Deploy Auth Inbox to Cloudflare Workers` → `Run workflow`.
 
-   2. **Deploy the Cloudflare Worker**
+   After success, **delete the workflow logs** to avoid leaking your config.
 
-      [![Deploy to Cloudflare Workers](https://deploy.workers.cloudflare.com/button)](https://deploy.workers.cloudflare.com/?url=https://github.com/TooonyChen/AuthInbox)
+3. Jump to [Set Email Forwarding](#3-set-email-forwarding-).
 
-      1. Click the button on the top to fork this repository or directly fork this repository.
+---
 
-      2. Open the repository that you fork, find the `Actions` page, find `Deploy Auth Inbox to Cloudflare Workers`, and click `enable workflow` to activate the workflows.
+### Option 2 — Deploy via CLI
 
-      3. Then, in the repository page, navigate to `Settings` -> `Secrets and variables` -> `Actions` -> `Repository secrets` and add the following secrets:
-         - `CLOUDFLARE_ACCOUNT_ID`: Cloudflare account ID.
-         - `CLOUDFLARE_API_TOKEN`: Cloudflare API Token.
-         - `TOML`: Configuration file, refer to [wrangler.toml with comments](https://github.com/TooonyChen/AuthInbox/blob/main/wrangler.toml.example). **Please use the [version without comments](https://github.com/TooonyChen/AuthInbox/blob/main/wrangler.toml.example.clear) in adding the secret in case of unknown errors.**
+1. **Clone and install**
 
-      4. Back to `Action` Page in your repository, find `Deploy Auth Inbox to Cloudflare Workers` and press `Run workflow` to deploy the worker. If you can't find the `Run workflow` button, you can trigger GitHub Actions automatically by making a small edit to the `readme.md` file in your repository.
+   ```bash
+   git clone https://github.com/TooonyChen/AuthInbox.git
+   cd AuthInbox
+   pnpm install
+   ```
 
-      5. After the deployment is successful, you can find the URL of your worker in the `Deploy Auth Inbox to Cloudflare Workers` workflow log.
+2. **Create D1 database**
 
-      6. Find the `delete all logs` button in the upper-right corner of the workflow logs page. Delete the logs to avoid data leakage.
+   ```bash
+   pnpm wrangler d1 create inbox-d1
+   pnpm wrangler d1 execute inbox-d1 --remote --file=./db/schema.sql
+   ```
 
-      7. Done! ✅ Jump to step 3: Set Email Forwarding.
+   Copy the `database_id` from the output.
 
+3. **Configure**
 
-2. **Install using command-line**
+   ```bash
+   cp wrangler.toml.example wrangler.toml
+   ```
 
-	1. **Initialization**
+   Edit `wrangler.toml` — at minimum fill in:
 
-		```bash
-		npm install wrangler -g # install wrangler
-		git clone https://github.com/TooonyChen/AuthInbox.git # clone the repository
-		cd AuthInbox # change directory
-		npm install # install dependencies
-		npm --prefix web install # install dashboard dependencies
-		```
+   ```toml
+   [vars]
+   FrontEndAdminID       = "your-username"
+   FrontEndAdminPassword = "your-password"
+   UseBark               = "false"
 
-	2. **create d1 database**
+   # AI provider — choose any compatible service
+   AI_BASE_URL    = "https://generativelanguage.googleapis.com/v1beta/openai"
+   AI_API_KEY     = "your-api-key"
+   AI_API_FORMAT  = "openai"
+   AI_MODEL       = "gemini-2.0-flash"
 
-		When you execute the [Wrangler](https://developers.cloudflare.com/workers/wrangler/get-started/) login command for the first time, you will be prompted to log in. Just follow the prompts.
+   [[d1_databases]]
+   binding       = "DB"
+   database_name = "inbox-d1"
+   database_id   = "<your-database-id>"
+   ```
 
-		```bash
-		npx wrangler d1 create inbox-d1 # creating a d1 database called 'inbox-d1'
-		npx wrangler d1 execute inbox-d1 --remote --file=./db/schema.sql # write the schema.sql to the database
-		```
-		you will get the result like this:
-		```bash
-		✅ Successfully created DB 'inbox-d1'
+   **`AI_API_FORMAT`** options:
 
-		[[d1_databases]]
-		binding = "DB" # available in your Worker on env.DB
-		database_name = "inbox-d1"
-		database_id = "<unique-ID-for-your-database>"
-		```
-		please copy the result from your terminal, you will use them in the next step
+   | Value | Endpoint | Compatible providers |
+   |---|---|---|
+   | `openai` | `/v1/chat/completions` | OpenAI, Gemini (OpenAI-compat), DeepSeek, Groq, Mistral, … |
+   | `responses` | `/v1/responses` | OpenAI Responses API |
+   | `anthropic` | `/v1/messages` | Anthropic Claude |
 
-	3. **Configure Environment Variables**
+   **Common `AI_BASE_URL` values:**
+   ```
+   OpenAI:              https://api.openai.com
+   Gemini (OAI-compat): https://generativelanguage.googleapis.com/v1beta/openai
+   Anthropic:           https://api.anthropic.com
+   DeepSeek:            https://api.deepseek.com
+   Groq:                https://api.groq.com/openai
+   ```
 
-		Use `wrangler.toml` file in the project root with the necessary environment variables:
+   **Optional fallback provider** (triggered if primary fails after 3 retries):
+   ```toml
+   # AI_FALLBACK_BASE_URL   = "https://api.openai.com"
+   # AI_FALLBACK_API_KEY    = "your-fallback-key"
+   # AI_FALLBACK_API_FORMAT = "openai"
+   # AI_FALLBACK_MODEL      = "gpt-4o-mini"
+   ```
 
-		```toml
-		name = "auth-inbox"
-		type = "typescript"
+   Optional Bark vars: `barkTokens`, `barkUrl`.
 
-		[vars]
-		UseBark = "true" # set 'true' to use or 'false' to not use
-		barkUrl = "https://api.day.app"
-		barkTokens = "[token1, token2]" # set to your bark tokens on your iOS device, download it from https://bark.day.app/, you can use multiple tokens, if you only use one, then set it to '[token1]'
-		FrontEndAdminID = "admin" # your login
-		FrontEndAdminPassword = "password" # your password
-		GoogleAPIKey = "xxxxxxxxxxx" # your google api, go to https://aistudio.google.com/ to generate one if u dont have
+4. **Build and deploy**
 
-		[[d1_databases]] # Copy the lines obtained from step 2 from your terminal.
-		binding = "DB"
-		database_name = "inbox-d1" # Copy from step 2
-		database_id = "<unique-ID-for-your-database>" # Copy from step 2
-		```
+   ```bash
+   pnpm run deploy
+   ```
 
-	4. **Deploy your own worker** 🌐
-	   Deploy your Worker to make your project accessible on the Internet. Run:
-	   ```bash
-	   npm run build:web
-	   npx wrangler deploy
-	   ```
-	   You will get output like this:
-	   ```
-	   Outputs: https://auth-inbox.<YOUR_SUBDOMAIN>.workers.dev
-	   ```
-	   You can now visit the URL for your newly depolyed Auth Inbox for checking the email results.
+   Output: `https://auth-inbox.<your-subdomain>.workers.dev`
 
-3. **Set Email Forwarding** ✉️
+---
 
-	Go to [Cloudflare Dashboard](https://dash.cloudflare.com/) -> `Websites` -> `<your-domain>` -> `Email` -> `Email-Routing` -> `Routing Rules`
+### 3. Set Email Forwarding ✉️
 
-	if you want to use `catch-all address`:
-	![image](https://github.com/user-attachments/assets/53e5a939-6b03-4ca6-826a-7a5f02f361ac)
+Go to [Cloudflare Dashboard](https://dash.cloudflare.com/) → `Websites` → `<your-domain>` → `Email` → `Email Routing` → `Routing Rules`.
 
-	if you want to use `custom address`:
-	![image](https://github.com/user-attachments/assets/b0d0ab94-c2ad-4870-ac08-d53e64b2c880)
+**Catch-all** (forwards all addresses to the Worker):
+![image](https://github.com/user-attachments/assets/53e5a939-6b03-4ca6-826a-7a5f02f361ac)
 
-4. **Enjoy!** 🎉
+**Custom address** (forwards a specific address):
+![image](https://github.com/user-attachments/assets/b0d0ab94-c2ad-4870-ac08-d53e64b2c880)
 
-	All set! ✅ You can now receive and view your authentication emails securely and efficiently using Auth Inbox!
+### 4. Done! 🎉
+
+Visit your Worker URL, log in with the credentials you set, and start receiving verification emails.
 
 ---
 
 ## License 📜
 
-This project is licensed under the [MIT License](LICENSE).
+[MIT License](LICENSE)
 
 ---
 
@@ -221,27 +192,26 @@ This project is licensed under the [MIT License](LICENSE).
 
 ![image](https://github.com/user-attachments/assets/ec14d226-ae82-4689-b44d-22850002c34c)
 
-
 ---
 
 ## Acknowledgements 🙏
 
-- **Cloudflare Workers** for providing a powerful serverless platform.
-- **Google Gemini AI** for enabling intelligent email content extraction.
-- **Bark** for real-time notification capabilities.
-- **Open Source Community** for inspiring and supporting projects like AuthInbox.
-- **ChatGPT** for helping me writing some of the code.
+- **Cloudflare Workers** for the serverless platform.
+- **Google Gemini AI** for intelligent email extraction.
+- **Bark** for real-time notifications.
+- **shadcn/ui** for the component library.
+- **Open Source Community** for the inspiration.
 
 ---
 
 ## TODO 📝
 
-- [x] **Github Pages Deployment**: Automatically deploy to Cloudflare Workers via Github Actions.
-- [ ] **Regular Expressions**: Using regular expressions instead of Google Gemini AI for privacy, make Gemini AI to an optional function.
-- [ ] **Multi-User Support**: Add functionality to manage multiple users for increased flexibility and broader usage.
-- [ ] **Enhance Front-End Design**: Improve the UI/UX of the web interface for a more modern and user-friendly experience.
-- [ ] **API**: Add API functionality.
-- [ ] **Sending Email**: Add the ability to send emails.
-- [ ] **More Notification Methods**: Support additional notification methods such as email, Slack, etc.
-
----
+- [x] GitHub Actions deployment
+- [x] OpenAI fallback support
+- [x] React dashboard with shadcn/ui
+- [x] Promotional email filter (skip ads before LLM)
+- [x] Raw email viewer + sandboxed HTML preview
+- [ ] Regex-based extraction as an AI-free option
+- [ ] Multi-user support
+- [ ] More notification methods (Slack, webhook, etc.)
+- [ ] Sending email support
diff --git a/README_CN.md b/README_CN.md
index 72ab042..25a19a5 100644
--- a/README_CN.md
+++ b/README_CN.md
@@ -2,9 +2,11 @@
 
 [English](https://github.com/TooonyChen/AuthInbox/blob/main/README.md) | [简体中文](https://github.com/TooonyChen/AuthInbox/blob/main/README_CN.md)
 
-**Auth Inbox** 是一个自建的开源多邮箱验证码的接码平台，基于 [Cloudflare](https://cloudflare.com/) 的免费服务。它可以自动处理收到的邮件，提取验证码或链接，并将其存储在数据库中。管理员可以通过一个用户友好的网页界面轻松查看提取的信息。AuthInbox 还支持通过 Bark 进行实时通知，使其成为一个全面且省心的邮件认证管理解决方案。
+**Auth Inbox** 是一个开源的自建验证码接码平台，基于 [Cloudflare](https://cloudflare.com/) 的免费无服务器服务。它会自动处理收到的邮件，在调用 AI 之前过滤掉广告邮件，提取验证码或链接并存入数据库。现代化的 React 仪表盘让管理员可以查看提取的验证码、查看原始邮件内容、渲染 HTML 邮件预览，全程受 Basic Auth 保护。
 
-不想在主邮箱中收到广告和垃圾邮件？想要多个备用邮箱用于注册服务和网站？试试这个吧！
+不想在主邮箱中收到广告和垃圾邮件？需要多个备用邮箱用于注册各类服务？试试这个**安全**、**无服务器**、**轻量**的解决方案吧！
+
+[![Deploy to Cloudflare Workers](https://deploy.workers.cloudflare.com/button)](https://deploy.workers.cloudflare.com/?url=https://github.com/TooonyChen/AuthInbox)
 
 ![框架](https://github.com/user-attachments/assets/f2ea1ef1-b500-4e47-88ba-a00d19f99595)
 
@@ -12,208 +14,177 @@
 
 ## 目录 📑
 
-- [功能](#features)
-- [使用的技术](#technologies-used)
-- [安装](#installation)
-- [许可证](#license)
-- [截图](#Screenshots)
+- [功能](#功能-)
+- [使用的技术](#使用的技术-)
+- [安装](#安装-)
+- [许可证](#许可证-)
+- [截图](#截图-)
 
 ---
 
 ## 功能 ✨
 
-- **邮件处理**：自动捕获和存储收到的邮件。
-- **验证码提取**：利用 AI 从邮件中提取验证码、链接和组织名称。
-- **安全前端**：提供受 Basic Access Authentication 保护的网页界面，用于查看提取的验证码。
-- **实时通知**：当提取到新的验证码时，可选通过 Bark 发送通知。
-- **数据库集成**：将原始和处理过的邮件数据存储在 D1Database 中。
+- **广告过滤**：通过邮件头（`List-Unsubscribe`、`Precedence: bulk` 等）识别并跳过营销邮件，不调用 AI，节省 token。
+- **AI 验证码提取**：使用 Google Gemini（OpenAI 作为备用）提取验证码、链接和组织名称。
+- **现代化仪表盘**：React 18 + shadcn/ui 界面，包含邮件列表、详情面板和三个 Tab —— 提取内容、原始邮件、HTML 渲染预览。
+- **安全 HTML 预览**：邮件 HTML 经 DOMPurify 净化后在沙箱 iframe 中渲染。
+- **一键复制**：验证码和链接均有复制按钮，复制后有 Toast 提示。
+- **实时通知**：可选接入 Bark，收到新验证码时推送到 iOS 设备。
+- **数据库存储**：所有原始邮件和 AI 提取结果均存入 Cloudflare D1。
 
 ---
 
 ## 使用的技术 🛠️
 
-- **Cloudflare Workers**: 无服务器平台，用于处理邮件处理和Web请求。
-- **Cloudflare D1**: Cloudflare的无服务器SQL数据库，用于存储邮件数据。
-- **TypeScript**: 强类型编程语言，确保代码的稳健性和可维护性。
-- **AI 提示词优化**: 定制的提示确保从多种邮件格式中精确提取标题、代码和主题。
-- **Google AI Studio API**: 利用优化的AI提示从邮件中提取相关信息，以提升数据的准确性和可靠性。
-- **Bark API**: 可选集成，用于发送实时通知。
-- **HTML/CSS**: 前端界面，具有响应式和现代化设计。
-- **Google Fonts**: 通过一致的排版增强Web界面的视觉吸引力。
-
-
----
-
-## AI 提示词优化 🧠
-
-为了确保从收到的电子邮件中准确提取信息，我们使用Google AI Studio API实施了AI提示优化。通过设计精确且具有上下文意识的提示，AI可以可靠地识别和提取关键要素，如：
-
-- **组织名称（标题）**: 识别发件人的组织或公司。
-- **验证码/链接**: 提取账户验证所需的代码、链接或密码。
-- **电子邮件主题**: 总结电子邮件的主要目的，例如“账户验证”或“密码重置”。
-
-**提示词如下:**
-```plaintext
-Email content: [Insert raw email content here].
-
-Please read the email and extract the following information:
-1. Code/Link/Password from the email (if available).
-2. Organization name (title) from which the email is sent.
-3. A brief summary of the email's topic (e.g., 'account verification').
-
-Format the output as JSON with this structure:
-{
-  "title": "The organization or company that sent the verification code (e.g., 'Netflix')",
-  "code": "The extracted verification code, link, or password (e.g., '123456' or 'https://example.com/verify?code=123456')",
-  "topic": "A brief summary of the email's topic (e.g., 'account verification')",
-  "codeExist": 1
-}
-
-If both a code and a link are present, include both in the 'code' field like this:
-"code": "code, link"
-
-If there is no code, clickable link, or this is an advertisement email, return:
-{
-  "codeExist": 0
-}
-
-```
+- **Cloudflare Workers** — 处理邮件和 API 的无服务器运行时。
+- **Cloudflare D1** — 兼容 SQLite 的无服务器数据库。
+- **Cloudflare Email Routing** — 将收到的邮件路由到 Worker。
+- **React 18 + Vite + Tailwind CSS + shadcn/ui** — 前端仪表盘。
+- **任意 OpenAI 兼容 / Anthropic AI 提供商** — 通过环境变量配置，支持 Gemini、OpenAI、DeepSeek、Groq、Anthropic 等。
+- **Bark API** — iOS 实时推送通知（可选）。
+- **TypeScript** — 端到端类型安全。
 
 ---
 
 ## 安装 ⚙️
-0. **先决条件**
 
-	- 创建一个 [Google AI Studio API](https://aistudio.google.com/)
+### 先决条件
 
-	- 在你的 [Cloudflare](https://dash.cloudflare.com/) 账户上绑定一个域名
+- 一个 [Google AI Studio API Key](https://aistudio.google.com/)
+- 在 [Cloudflare](https://dash.cloudflare.com/) 账户上绑定一个域名
+- Cloudflare 账户 ID 和 API Token，可在 [此处](https://dash.cloudflare.com/profile/api-tokens) 获取
+- *（可选）* [Bark App](https://bark.day.app/)，用于 iOS 推送通知
 
-	  - 获取你的Cloudflare 账户 ID, 可在 [Cloudflare API Tokens](https://dash.cloudflare.com/profile/api-tokens) 中找到
+---
 
-      - 获取你的 Cloudflare API Token，可在 [Cloudflare API Tokens](https://dash.cloudflare.com/profile/api-tokens) 中找到
+### 方式一 — 通过 GitHub Actions 部署
 
-	- （可选）下载[Bark App](https://bark.day.app/)，在App中获得一个Bark Token
+1. **创建 D1 数据库**
 
-1. **使用 Github Pages 进行安装**
+   进入 [Cloudflare 仪表盘](https://dash.cloudflare.com/) → `Workers & Pages` → `D1 SQL Database` → `Create`，名称填 `inbox-d1`。
 
-	1. **创建 D1 数据库**
+   点击进入数据库 → `Console`，粘贴并执行 [`db/schema.sql`](https://github.com/TooonyChen/AuthInbox/blob/main/db/schema.sql) 中的内容。
 
-		1. 进入 [Cloudflare 仪表盘](https://dash.cloudflare.com/) -> `Workers & Pages` -> `D1 SQL Database` -> `Create`
+   复制 `database_id`，下一步会用到。
 
-		2. 输入名称 `inbox-d1` 并点击 `Create`
+2. **Fork 并部署**
 
-		3. 创建 `inbox-d1` 后，点击进入并找到 `Console`
+   [![Deploy to Cloudflare Workers](https://deploy.workers.cloudflare.com/button)](https://deploy.workers.cloudflare.com/?url=https://github.com/TooonyChen/AuthInbox)
 
-		4. 在控制台中执行 [db/schema.sql](https://github.com/TooonyChen/AuthInbox/blob/main/db/schema.sql) 中的 SQL 命令，直接复制、粘贴并执行它。
+   在你 Fork 的仓库中，进入 `Settings` → `Secrets and variables` → `Actions`，添加以下 Secrets：
+   - `CLOUDFLARE_ACCOUNT_ID`
+   - `CLOUDFLARE_API_TOKEN`
+   - `TOML` — 使用[不带注释的模板](https://github.com/TooonyChen/AuthInbox/blob/main/wrangler.toml.example.clear)，避免解析报错。
 
-		5. 复制 `database_id` 和 `database_name`，用于下一步配置 `TOML` 文件时使用
+   然后进入 `Actions` → `Deploy Auth Inbox to Cloudflare Workers` → `Run workflow`。
 
-	2. **部署 Cloudflare Worker**
+   部署成功后，**务必删除 workflow 日志**，防止配置信息泄露。
 
-		[![Deploy to Cloudflare Workers](https://deploy.workers.cloudflare.com/button)](https://deploy.workers.cloudflare.com/?url=https://github.com/TooonyChen/AuthInbox)
+3. 跳转到[设置邮件转发](#3-设置邮件转发-)。
 
-		1. 点击上方按钮 fork 此仓库，或直接 fork 此仓库。
+---
 
-		2. 打开你 fork 的仓库，找到 `Actions` 页面，找到 `Deploy Auth Inbox to Cloudflare Workers`，并点击 `enable workflow` 激活 workflow。
+### 方式二 — 通过命令行部署
 
-		3. 然后，在仓库页面中，导航到 `Settings` -> `Secrets and variables` -> `Actions` -> `Repository secrets`，并添加以下 secrets：
-			- `CLOUDFLARE_ACCOUNT_ID`: Cloudflare 账户 ID。
-			- `CLOUDFLARE_API_TOKEN`: Cloudflare API Token。
-			- `TOML`: 配置文件，参考 [带有注释的wrangler.toml](https://github.com/TooonyChen/AuthInbox/blob/main/wrangler.toml.example)。**添加至secrets时，请使用[不带注释的版本](https://github.com/TooonyChen/AuthInbox/blob/main/wrangler.toml.example.clear)来避免奇怪的报错。**
+1. **克隆并安装依赖**
 
-		4. 返回你仓库的 `Actions` 页面，找到 `Deploy Auth Inbox to Cloudflare Workers`，并按 `Run workflow` 来部署 worker。如果没找到 `Run workflow` 按钮，请你随便修改一下repo中的`readme.md`文件，让Github Actions自动运行。
+   ```bash
+   git clone https://github.com/TooonyChen/AuthInbox.git
+   cd AuthInbox
+   pnpm install
+   ```
 
-		5. 部署成功后，你可以在 `Deploy Auth Inbox to Cloudflare Workers` 的 workflow 日志中找到你的 worker URL。
+2. **创建 D1 数据库**
 
-		6. 在 workflow 日志页面的右上角找到 `delete all logs` 来删除日志，删除它来避免数据外泄。
+   ```bash
+   pnpm wrangler d1 create inbox-d1
+   pnpm wrangler d1 execute inbox-d1 --remote --file=./db/schema.sql
+   ```
 
-		7. 完成！✅ 请前往第三步：设置邮件转发。
+   复制输出中的 `database_id`。
 
-2. **使用 Wrangler 命令行部署到 Cloudflare Workers**
+3. **配置**
 
-   1. **初始化**
+   ```bash
+   cp wrangler.toml.example wrangler.toml
+   ```
 
-      ```bash
-	      npm install wrangler -g # 安装 wrangler
-	      git clone https://github.com/TooonyChen/AuthInbox.git # 克隆仓库
-	      cd AuthInbox # 切换目录
-	      npm install # 安装依赖
-	      npm --prefix web install # 安装前端仪表盘依赖
-	      ```
-
-   2. **创建 d1 数据库**
-
-      当你第一次执行 [Wrangler](https://developers.cloudflare.com/workers/wrangler/get-started/) 登录命令时，系统会提示你登录。按提示操作即可。
-
-      ```bash
-      npx wrangler d1 create inbox-d1 # 创建名为 'inbox-d1' 的 d1 数据库
-      npx wrangler d1 execute inbox-d1 --remote --file=./db/schema.sql # 执行 schema.sql 文件
-      ```
-      你将会看到如下结果：
-      ```bash
-      ✅ Successfully created DB 'inbox-d1'
-
-      [[d1_databases]]
-      binding = "DB" # 在你的 Worker 中通过 env.DB 访问
-      database_name = "inbox-d1"
-      database_id = "<你的数据库的唯一ID>"
-      ```
-      请从终端复制结果，你将在下一步中使用它们。
-
-   3. **配置环境变量**
-
-   使用项目根目录下的 `wrangler.toml` 文件，并添加所需的环境变量：
+   编辑 `wrangler.toml`，至少填写以下内容：
 
    ```toml
-   name = "auth-inbox"
-   type = "typescript"
-
    [vars]
-   UseBark = 'true' # 设置为 'true' 启用 Bark，设置为 'false' 禁用
-   barkUrl = "https://api.day.app"
-   barkTokens = "[token1, token2]" # 填写你 iOS 设备上的 Bark tokens，可从 https://bark.day.app/ 下载应用获取，可以填写多个。如果你只想用一个，那么填写 '[token1]'
-   FrontEndAdminID = "admin" # 你的登录 ID
-   FrontEndAdminPassword = "password" # 你的登录密码
-   GoogleAPIKey = "xxxxxxxxxxx" # 你的 Google API key，如果没有可以前往 https://aistudio.google.com/ 生成一个
+   FrontEndAdminID       = "你的登录账号"
+   FrontEndAdminPassword = "你的登录密码"
+   UseBark               = "false"
 
-   [[d1_databases]] # 从步骤 2 的终端结果中复制这些行。
-   binding = "DB"
-   database_name = "inbox-d1" # 从步骤 2 中复制
-   database_id = "<你的数据库的唯一ID>" # 从步骤 2 中复制
+   # AI 提供商配置
+   AI_BASE_URL    = "https://generativelanguage.googleapis.com/v1beta/openai"
+   AI_API_KEY     = "你的 API Key"
+   AI_API_FORMAT  = "openai"
+   AI_MODEL       = "gemini-2.0-flash"
+
+   [[d1_databases]]
+   binding       = "DB"
+   database_name = "inbox-d1"
+   database_id   = "<你的数据库ID>"
    ```
 
-   4. **部署你的 worker** 🌐
+   **`AI_API_FORMAT`** 三选一：
 
-	   部署你的 Worker 以使项目在互联网上可访问。运行以下命令：
-	   ```bash
-	   npm run build:web
-	   npx wrangler deploy
-	   ```
-   你将看到如下输出：
+   | 值 | 实际请求路径 | 适用提供商 |
+   |---|---|---|
+   | `openai` | `/v1/chat/completions` | OpenAI、Gemini（OpenAI 兼容）、DeepSeek、Groq、Mistral 等 |
+   | `responses` | `/v1/responses` | OpenAI Responses API |
+   | `anthropic` | `/v1/messages` | Anthropic Claude 直连 |
+
+   **常用 `AI_BASE_URL`：**
    ```
-   output: https://auth-inbox.<你的子域名>.workers.dev
+   OpenAI:                https://api.openai.com
+   Gemini（OpenAI 兼容）: https://generativelanguage.googleapis.com/v1beta/openai
+   Anthropic:             https://api.anthropic.com
+   DeepSeek:              https://api.deepseek.com
+   Groq:                  https://api.groq.com/openai
    ```
-   你现在可以访问该 URL 来查看你部署的 Auth Inbox 的邮件面板。
 
-3. **设置邮件转发** ✉️
+   **备用模型（可选）**，主模型失败重试 3 次后触发：
+   ```toml
+   # AI_FALLBACK_BASE_URL   = "https://api.openai.com"
+   # AI_FALLBACK_API_KEY    = "备用 API Key"
+   # AI_FALLBACK_API_FORMAT = "openai"
+   # AI_FALLBACK_MODEL      = "gpt-4o-mini"
+   ```
 
-   前往 [Cloudflare Dashboard](https://dash.cloudflare.com/) -> `Websites` -> `<你的域名>` -> `Email` -> `Email-Routing` -> `Routing Rules`
+   可选 Bark 配置：`barkTokens`、`barkUrl`。
 
-   如果你想使用“接收所有地址”：
-   ![image](https://github.com/user-attachments/assets/53e5a939-6b03-4ca6-826a-7a5f02f361ac)
+4. **构建并部署**
 
-   如果你想使用“自定义地址”：
-   ![image](https://github.com/user-attachments/assets/b0d0ab94-c2ad-4870-ac08-d53e64b2c880)
+   ```bash
+   pnpm run deploy
+   ```
 
-4. **完成！** ✅
+   输出：`https://auth-inbox.<你的子域名>.workers.dev`
 
-   现在你可以访问你的新部署的 Auth Inbox URL，查看电子邮件结果。
+---
+
+### 3. 设置邮件转发 ✉️
+
+进入 [Cloudflare 仪表盘](https://dash.cloudflare.com/) → `Websites` → `<你的域名>` → `Email` → `Email Routing` → `Routing Rules`。
+
+**接收所有地址**（将所有邮件转发给 Worker）：
+![image](https://github.com/user-attachments/assets/53e5a939-6b03-4ca6-826a-7a5f02f361ac)
+
+**自定义地址**（转发特定地址）：
+![image](https://github.com/user-attachments/assets/b0d0ab94-c2ad-4870-ac08-d53e64b2c880)
+
+### 4. 完成！🎉
+
+访问你的 Worker URL，用你设置的账号密码登录，即可开始接收验证邮件。
 
 ---
 
 ## 许可证 📜
 
-本项目基于 [MIT License](LICENSE) 许可证。
+[MIT License](LICENSE)
 
 ---
 
@@ -221,24 +192,26 @@ If there is no code, clickable link, or this is an advertisement email, return:
 
 ![image](https://github.com/user-attachments/assets/41db550c-8340-4315-ba87-85330abc5dfb)
 
-
 ---
 
 ## 鸣谢 🙏
 
-- 感谢 **Cloudflare Workers** 提供强大的无服务器平台。
-- 感谢 **Google Gemini AI** 提供智能的邮件内容提取功能。
-- 感谢 **Bark** 提供实时通知能力。
-- 感谢 **开源社区** 为像 Auth Inbox 这样的项目提供灵感和支持。
-- 感谢 **ChatGPT** 协助我写代码
+- **Cloudflare Workers** 提供强大的无服务器平台。
+- **Google Gemini AI** 提供智能邮件内容提取能力。
+- **Bark** 提供实时推送通知。
+- **shadcn/ui** 提供组件库。
+- **开源社区** 提供灵感与支持。
 
 ---
+
 ## TODO 📝
 
-- [x] **Github Pages 部署**：通过 Github Actions 自动部署到 Cloudflare Workers。
-- [ ] **正则表达式**：使用正则表达式替代 Google Gemini AI 以保护隐私，并将使用 AI 作为一个可选功能。
-- [ ] **多用户支持**：增加管理多用户的功能，以提高灵活性并扩展使用范围。
-- [ ] **增强前端设计**：改进网页界面的UI/UX设计，使其更加现代化和用户友好。
-- [ ] **API**: 添加API接口。
-- [ ] **发送邮件**：添加发送邮件功能。
-- [ ] **更多通知方式**：支持更多通知方式，如邮件、Slack等。
+- [x] GitHub Actions 自动部署
+- [x] OpenAI 备用模型支持
+- [x] React 仪表盘（shadcn/ui）
+- [x] 广告邮件过滤（调用 AI 前拦截）
+- [x] 原始邮件查看 + 沙箱 HTML 预览
+- [ ] 正则表达式提取（无 AI 选项）
+- [ ] 多用户支持
+- [ ] 更多通知方式（Slack、Webhook 等）
+- [ ] 发送邮件功能
diff --git a/src/index.ts b/src/index.ts
index 44c62f5..0932d6e 100644
--- a/src/index.ts
+++ b/src/index.ts
@@ -10,34 +10,35 @@ import { RPCEmailMessage } from "./rpcEmail";
 
 import indexHtml from "./index.html";
 
+type ApiFormat = "openai" | "responses" | "anthropic";
+
+interface ProviderConfig {
+  baseUrl: string;
+  apiKey: string;
+  format: ApiFormat;
+  model: string;
+}
+
 interface Env {
-  // If you set another name in wrangler.toml as the value for 'binding',
-  // replace "DB" with the variable name you defined.
   DB: D1Database;
   ASSETS?: Fetcher;
   FrontEndAdminID: string;
   FrontEndAdminPassword: string;
+  UseBark: string;
   barkTokens: string;
   barkUrl: string;
-  GoogleAPIKey: string;
-  UseBark: string;
-  GeminiModel?: string;
-  OpenAIAPIKey?: string;
-  OpenAIBaseUrl?: string;
-  OpenAIModel?: string;
+  // Primary AI provider
+  AI_BASE_URL: string;
+  AI_API_KEY: string;
+  AI_API_FORMAT: ApiFormat;
+  AI_MODEL: string;
+  // Fallback AI provider (all four must be set to enable fallback)
+  AI_FALLBACK_BASE_URL?: string;
+  AI_FALLBACK_API_KEY?: string;
+  AI_FALLBACK_API_FORMAT?: ApiFormat;
+  AI_FALLBACK_MODEL?: string;
 }
 
-interface ModelResponse {
-  ok: boolean;
-  status: number;
-  statusText: string;
-  payload: unknown;
-}
-
-const DEFAULT_GEMINI_MODEL = "gemini-2.0-flash";
-const DEFAULT_OPENAI_BASE_URL = "https://api.openai.com";
-const DEFAULT_OPENAI_MODEL = "gpt-4o-mini";
-
 // Normalize model output into JSON text // 将模型输出规范化为可解析的 JSON 文本
 function extractJsonFromText(rawText: string): Record<string, unknown> | null {
   let candidate = rawText.trim();
@@ -58,49 +59,98 @@ function extractJsonFromText(rawText: string): Record<string, unknown> | null {
   }
 }
 
-// Read structured content from Gemini response // 读取 Gemini 响应中的结构化文本
-function extractGeminiText(payload: unknown): string | null {
-  const candidate = (payload as any)?.candidates?.[0]?.content?.parts?.[0];
-  if (!candidate || typeof candidate.text !== "string") {
-    console.error("Gemini response is missing expected data structure");
-    return null;
-  }
-  return candidate.text;
-}
-
-// Read structured content from OpenAI response // 读取 OpenAI 响应中的结构化文本
-function extractOpenAIText(payload: unknown): string | null {
-  const message = (payload as any)?.choices?.[0]?.message;
-  if (!message) {
-    console.error("OpenAI response is missing choices");
-    return null;
-  }
-
-  const content = message.content;
-  if (typeof content === "string") {
-    return content;
-  }
-
-  if (Array.isArray(content)) {
-    const textPart = content.find((part: any) => part?.type === "text" && typeof part.text === "string");
-    if (textPart) {
-      return textPart.text;
-    }
-  }
-
-  if (typeof message?.content?.[0]?.text === "string") {
-    return message.content[0].text;
-  }
-
-  console.error("OpenAI response is missing text content");
-  return null;
-}
-
 // Remove trailing slashes to avoid double separators // 移除结尾斜杠防止重复拼接
 function normaliseBaseUrl(baseUrl: string): string {
   return baseUrl.replace(/\/+$/, "");
 }
 
+// Unified AI provider caller supporting openai / responses / anthropic formats
+async function callProvider(config: ProviderConfig, prompt: string): Promise<string | null> {
+  const base = normaliseBaseUrl(config.baseUrl);
+  let endpoint: string;
+  let body: unknown;
+  const headers: Record<string, string> = { "Content-Type": "application/json" };
+
+  if (config.format === "openai") {
+    endpoint = `${base}/v1/chat/completions`;
+    headers["Authorization"] = `Bearer ${config.apiKey}`;
+    body = {
+      model: config.model,
+      temperature: 0,
+      response_format: { type: "json_object" },
+      messages: [
+        { role: "system", content: "You return only valid JSON that matches the requested schema." },
+        { role: "user", content: prompt },
+      ],
+    };
+  } else if (config.format === "responses") {
+    endpoint = `${base}/v1/responses`;
+    headers["Authorization"] = `Bearer ${config.apiKey}`;
+    body = {
+      model: config.model,
+      input: [{ role: "user", content: prompt }],
+      text: { format: { type: "json_object" } },
+    };
+  } else {
+    // anthropic
+    endpoint = `${base}/v1/messages`;
+    headers["x-api-key"] = config.apiKey;
+    headers["anthropic-version"] = "2023-06-01";
+    body = {
+      model: config.model,
+      max_tokens: 1024,
+      messages: [{ role: "user", content: prompt }],
+    };
+  }
+
+  let response: Response;
+  try {
+    response = await fetch(endpoint, { method: "POST", headers, body: JSON.stringify(body) });
+  } catch (err) {
+    console.error(`[callProvider:${config.format}] fetch error:`, err);
+    return null;
+  }
+
+  if (!response.ok) {
+    console.error(`[callProvider:${config.format}] HTTP ${response.status} ${response.statusText}`);
+    return null;
+  }
+
+  let payload: unknown;
+  try {
+    payload = await response.json();
+  } catch {
+    console.error(`[callProvider:${config.format}] failed to parse JSON response`);
+    return null;
+  }
+
+  // Extract text from response based on format
+  if (config.format === "openai") {
+    const content = (payload as any)?.choices?.[0]?.message?.content;
+    if (typeof content === "string") return content;
+    if (Array.isArray(content)) {
+      const part = content.find((p: any) => p?.type === "text" && typeof p.text === "string");
+      return part?.text ?? null;
+    }
+    console.error("[callProvider:openai] unexpected response shape");
+    return null;
+  }
+
+  if (config.format === "responses") {
+    const text = (payload as any)?.output?.[0]?.content?.[0]?.text
+      ?? (payload as any)?.output?.[0]?.text;
+    if (typeof text === "string") return text;
+    console.error("[callProvider:responses] unexpected response shape");
+    return null;
+  }
+
+  // anthropic
+  const text = (payload as any)?.content?.[0]?.text;
+  if (typeof text === "string") return text;
+  console.error("[callProvider:anthropic] unexpected response shape");
+  return null;
+}
+
 function jsonResponse(data: unknown, status = 200): Response {
   return new Response(JSON.stringify(data), {
     status,
@@ -406,113 +456,27 @@ export default class extends WorkerEntrypoint<Env> {
     }
   }
 
-  // Call Gemini for primary extraction // 调用 Gemini 执行主解析
-  private async callGemini(prompt: string, apiKey: string, model: string): Promise<ModelResponse> {
-    const endpoint = `https://generativelanguage.googleapis.com/v1beta/models/${model}:generateContent?key=${apiKey}`;
-    const init: RequestInit = {
-      method: "POST",
-      headers: {
-        "Content-Type": "application/json",
-      },
-      body: JSON.stringify({
-        contents: [
-          {
-            parts: [{ text: prompt }],
-          },
-        ],
-      }),
-    };
-
-    try {
-      const response = await fetch(endpoint, init);
-      let payload: unknown = null;
-      try {
-        payload = await response.json();
-      } catch (parseError) {
-        console.error("Gemini JSON parse error:", parseError);
-      }
-      return {
-        ok: response.ok,
-        status: response.status,
-        statusText: response.statusText,
-        payload,
-      };
-    } catch (error) {
-      console.error("Gemini request threw:", error);
-      return {
-        ok: false,
-        status: 0,
-        statusText: "FETCH_ERROR",
-        payload: null,
-      };
-    }
-  }
-
-  // Call OpenAI as the fallback extractor // 调用 OpenAI 作为兜底解析
-  private async callOpenAI(
-    prompt: string,
-    baseUrl: string,
-    apiKey: string,
-    model: string
-  ): Promise<ModelResponse> {
-    const endpoint = `${normaliseBaseUrl(baseUrl)}/v1/chat/completions`;
-    const init: RequestInit = {
-      method: "POST",
-      headers: {
-        "Content-Type": "application/json",
-        Authorization: `Bearer ${apiKey}`,
-      },
-      body: JSON.stringify({
-        model,
-        temperature: 0,
-        response_format: { type: "json_object" },
-        messages: [
-          {
-            role: "system",
-            content: "You return only valid JSON that matches the requested schema.",
-          },
-          {
-            role: "user",
-            content: prompt,
-          },
-        ],
-      }),
-    };
-
-    try {
-      const response = await fetch(endpoint, init);
-      let payload: unknown = null;
-      try {
-        payload = await response.json();
-      } catch (parseError) {
-        console.error("OpenAI JSON parse error:", parseError);
-      }
-      return {
-        ok: response.ok,
-        status: response.status,
-        statusText: response.statusText,
-        payload,
-      };
-    } catch (error) {
-      console.error("OpenAI request threw:", error);
-      return {
-        ok: false,
-        status: 0,
-        statusText: "FETCH_ERROR",
-        payload: null,
-      };
-    }
-  }
-
   // Primary email handler // 主要邮件处理入口
   async email(message: ForwardableEmailMessage): Promise<void> {
     const env: Env = this.env;
     const useBark = env.UseBark.toLowerCase() === "true";
-    const googleApiKey = env.GoogleAPIKey;
-    const geminiModel = env.GeminiModel || DEFAULT_GEMINI_MODEL;
-    const openAIKey = env.OpenAIAPIKey;
-    const openAIBaseUrl = env.OpenAIBaseUrl || DEFAULT_OPENAI_BASE_URL;
-    const openAIModel = env.OpenAIModel || DEFAULT_OPENAI_MODEL;
+
+    const primary: ProviderConfig = {
+      baseUrl: env.AI_BASE_URL,
+      apiKey: env.AI_API_KEY,
+      format: env.AI_API_FORMAT ?? "openai",
+      model: env.AI_MODEL,
+    };
+
+    const fallback: ProviderConfig | null =
+      env.AI_FALLBACK_BASE_URL && env.AI_FALLBACK_API_KEY && env.AI_FALLBACK_MODEL
+        ? {
+            baseUrl: env.AI_FALLBACK_BASE_URL,
+            apiKey: env.AI_FALLBACK_API_KEY,
+            format: env.AI_FALLBACK_API_FORMAT ?? "openai",
+            model: env.AI_FALLBACK_MODEL,
+          }
+        : null;
 
     // Pull raw email content // 获取原始邮件内容
     const rawEmail =
@@ -569,63 +533,41 @@ export default class extends WorkerEntrypoint<Env> {
 
     try {
       const maxRetries = 3;
-      let retryCount = 0;
       let extractedData: Record<string, unknown> | null = null;
-      let geminiStatusError = false;
 
-      // First attempt Gemini with simple retries // 先使用 Gemini 并进行简单重试
-      while (retryCount < maxRetries && !extractedData && !geminiStatusError) {
-        const geminiResult = await this.callGemini(aiPrompt, googleApiKey, geminiModel);
-        console.log(`Gemini response attempt ${retryCount + 1}:`, geminiResult.payload);
-
-        if (!geminiResult.ok) {
-          geminiStatusError = true;
-          console.error(
-            `Gemini request failed with status ${geminiResult.status} ${geminiResult.statusText}`
-          );
-          break;
-        }
-
-        const extractedText = extractGeminiText(geminiResult.payload);
-        if (extractedText) {
-          console.log(`Extracted Text before parsing: "${extractedText}"`);
-          const parsed = extractJsonFromText(extractedText);
+      // Primary provider: up to 3 attempts
+      for (let attempt = 1; attempt <= maxRetries; attempt++) {
+        console.log(`[primary:${primary.format}] attempt ${attempt}`);
+        const text = await callProvider(primary, aiPrompt);
+        if (text) {
+          const parsed = extractJsonFromText(text);
           if (parsed) {
             extractedData = parsed;
-            console.log("Parsed Extracted Data:", extractedData);
+            console.log("[primary] extracted data:", extractedData);
             break;
           }
         }
-
-        retryCount += 1;
-        if (retryCount < maxRetries) {
-          console.log("Retrying AI request...");
-        } else {
-          console.error("Max retries reached. Unable to get valid AI response.");
-        }
+        if (attempt < maxRetries) console.log("[primary] retrying...");
+        else console.error("[primary] max retries reached");
       }
 
-      if (!extractedData && geminiStatusError && openAIKey) {
-        // Gemini failed with non-200, switch to OpenAI fallback // 当 Gemini 返回非 200 时切换到 OpenAI 兜底
-        const openAIResult = await this.callOpenAI(aiPrompt, openAIBaseUrl, openAIKey, openAIModel);
-        console.log("OpenAI fallback response:", openAIResult.payload);
-        if (!openAIResult.ok) {
-          console.error(
-            `OpenAI fallback failed with status ${openAIResult.status} ${openAIResult.statusText}`
-          );
-        } else {
-          const openAIText = extractOpenAIText(openAIResult.payload);
-          if (openAIText) {
-            console.log(`OpenAI text before parsing: "${openAIText}"`);
-            const parsed = extractJsonFromText(openAIText);
-            if (parsed) {
-              extractedData = parsed;
-              console.log("Parsed Extracted Data from OpenAI:", extractedData);
-            }
+      // Fallback provider: one attempt if primary produced nothing and fallback is configured
+      if (!extractedData && fallback) {
+        console.log(`[fallback:${fallback.format}] attempting`);
+        const text = await callProvider(fallback, aiPrompt);
+        if (text) {
+          const parsed = extractJsonFromText(text);
+          if (parsed) {
+            extractedData = parsed;
+            console.log("[fallback] extracted data:", extractedData);
+          } else {
+            console.error("[fallback] failed to parse response");
           }
+        } else {
+          console.error("[fallback] provider returned nothing");
         }
-      } else if (!extractedData && geminiStatusError) {
-        console.error("Gemini request failed and OpenAI fallback is not configured.");
+      } else if (!extractedData) {
+        console.error("[primary] failed and no fallback configured");
       }
 
       if (extractedData) {
diff --git a/wrangler.toml.example b/wrangler.toml.example
index 76cc10d..9f7cf6f 100644
--- a/wrangler.toml.example
+++ b/wrangler.toml.example
@@ -57,30 +57,36 @@ FrontEndAdminPassword = "password"
 # 你的登录密码 **请更改为你自己的登录密码**
 
 
-GoogleAPIKey = "xxxxxxxxxxx"
-# GoogleAPIKey: String
-# your google AI API, you can have one from https://aistudio.google.com
-# 你的谷歌AI API，你可以从 https://aistudio.google.com 获取
+# --- Primary AI provider ---
+AI_BASE_URL = "https://generativelanguage.googleapis.com/v1beta/openai"
+# Base URL of your AI provider (no trailing slash).
+# Examples:
+#   OpenAI:            https://api.openai.com
+#   Gemini (OAI-compat): https://generativelanguage.googleapis.com/v1beta/openai
+#   Anthropic:         https://api.anthropic.com
+#   DeepSeek:          https://api.deepseek.com
+#   Groq:              https://api.groq.com/openai
 
-GeminiModel = "gemini-2.0-flash"
-# GeminiModel: String
-# optional override for the Gemini model, defaults to "gemini-2.0-flash"
-# 可选配置，覆盖默认的 Gemini 模型名称，默认值为 "gemini-2.0-flash"
+AI_API_KEY = "your-api-key"
+# API key for the primary provider. Use a Cloudflare Secret for production.
+# 主模型的 API Key，生产环境请使用 Cloudflare Secret 存储
 
-OpenAIAPIKey = ""
-# OpenAIAPIKey: String
-# optional key for OpenAI fallback when Gemini fails with a non-200 response
-# 可选配置，当 Gemini 返回非 200 时使用 OpenAI 作为兜底
+AI_API_FORMAT = "openai"
+# Request/response format. One of:
+#   openai     — POST /v1/chat/completions  (OpenAI, Gemini compat, DeepSeek, Groq, etc.)
+#   responses  — POST /v1/responses         (OpenAI Responses API)
+#   anthropic  — POST /v1/messages          (Anthropic Claude direct)
+# 请求格式，三选一：openai | responses | anthropic
 
-OpenAIBaseUrl = "https://api.openai.com"
-# OpenAIBaseUrl: String
-# optional, change only if you proxy the OpenAI API
-# 可选配置，仅在自建代理时更改
+AI_MODEL = "gemini-2.0-flash"
+# Model ID to use, e.g. "gpt-4o-mini", "gemini-2.0-flash", "claude-3-5-haiku-20241022"
+# 模型 ID
 
-OpenAIModel = "gpt-4o-mini"
-# OpenAIModel: String
-# optional override for the OpenAI model, defaults to "gpt-4o-mini"
-# 可选配置，覆盖默认的 OpenAI 模型
+# --- Fallback AI provider (optional — remove all four lines to disable) ---
+# AI_FALLBACK_BASE_URL   = "https://api.openai.com"
+# AI_FALLBACK_API_KEY    = "your-fallback-api-key"
+# AI_FALLBACK_API_FORMAT = "openai"
+# AI_FALLBACK_MODEL      = "gpt-4o-mini"
 
 # Bind a D1 database. D1 is Cloudflare’s native serverless SQL database.
 # Docs: https://developers.cloudflare.com/workers/wrangler/configuration/#d1-databases
diff --git a/wrangler.toml.example.clear b/wrangler.toml.example.clear
index 44d839d..c590d85 100644
--- a/wrangler.toml.example.clear
+++ b/wrangler.toml.example.clear
@@ -16,11 +16,14 @@ barkUrl = "https://api.day.app"
 barkTokens = "[token1, token2]"
 FrontEndAdminID = "admin"
 FrontEndAdminPassword = "password"
-GoogleAPIKey = "xxxxxxxxxxx"
-GeminiModel = "gemini-2.0-flash"
-OpenAIAPIKey = ""
-OpenAIBaseUrl = "https://api.openai.com"
-OpenAIModel = "gpt-4o-mini"
+AI_BASE_URL = "https://generativelanguage.googleapis.com/v1beta/openai"
+AI_API_KEY = "your-api-key"
+AI_API_FORMAT = "openai"
+AI_MODEL = "gemini-2.0-flash"
+# AI_FALLBACK_BASE_URL = "https://api.openai.com"
+# AI_FALLBACK_API_KEY = "your-fallback-api-key"
+# AI_FALLBACK_API_FORMAT = "openai"
+# AI_FALLBACK_MODEL = "gpt-4o-mini"
 
 [[d1_databases]]
 binding = "DB"