Revise preview feature stability note

Updated the preview feature note for stability.

.github/CONTRIBUTING.md

@@ -0,0 +1,35 @@
+# Contributing
+
+## Setup
+
+```bash
+git clone https://github.com/YOUR_USERNAME/next-ai-draw-io.git
+cd next-ai-draw-io
+npm install
+cp env.example .env.local
+npm run dev
+```
+
+## Code Style
+
+We use [Biome](https://biomejs.dev/) for linting and formatting:
+
+```bash
+npm run format   # Format code
+npm run lint     # Check lint errors
+npm run check    # Run all checks (CI)
+```
+
+Pre-commit hooks via Husky will run Biome automatically on staged files.
+
+For a better experience, install the [Biome VS Code extension](https://marketplace.visualstudio.com/items?itemName=biomejs.biome) for real-time linting and format-on-save.
+
+## Pull Requests
+
+1. Create a feature branch
+2. Make changes and ensure `npm run check` passes
+3. Submit PR against `main` with a clear description
+
+## Issues
+
+Include steps to reproduce, expected vs actual behavior, and AI provider used.

.github/ISSUE_TEMPLATE/bug_report.md

@@ -0,0 +1,35 @@
+---
+name: Bug Report
+about: Report a bug to help us improve
+title: '[Bug] '
+labels: bug
+assignees: ''
+---
+
+> **Note**: This template is just a guide. Feel free to ignore the format entirely - any feedback is welcome! Don't let the template stop you from sharing your thoughts.
+
+## Bug Description
+A brief description of the issue.
+
+## Steps to Reproduce
+1. Go to '...'
+2. Click on '...'
+3. Scroll to '...'
+4. See error
+
+## Expected Behavior
+What you expected to happen.
+
+## Actual Behavior
+What actually happened.
+
+## Screenshots
+If applicable, add screenshots to help explain the problem.
+
+## Environment
+- OS: [e.g. Windows 11, macOS 14]
+- Browser: [e.g. Chrome 120, Safari 17]
+- Version: [e.g. 1.0.0]
+
+## Additional Context
+Any other information about the problem.

.github/ISSUE_TEMPLATE/config.yml

@@ -0,0 +1,5 @@
+blank_issues_enabled: true
+contact_links:
+  - name: Discussions
+    url: https://github.com/DayuanJiang/next-ai-draw-io/discussions
+    about: Have questions or ideas? Feel free to start a discussion

.github/ISSUE_TEMPLATE/feature_request.md

@@ -0,0 +1,25 @@
+---
+name: Feature Request
+about: Suggest a new feature for this project
+title: '[Feature] '
+labels: enhancement
+assignees: ''
+---
+
+> **Note**: This template is just a guide. Feel free to ignore the format entirely - any feedback is welcome! Don't let the template stop you from sharing your ideas.
+
+## Feature Description
+A brief description of the feature you'd like.
+
+## Problem Context
+Is this related to a problem? Please describe.
+e.g. I'm always frustrated when [...]
+
+## Proposed Solution
+How you'd like this feature to work.
+
+## Alternatives Considered
+Any alternative solutions or features you've considered.
+
+## Additional Context
+Any other information or screenshots about the feature request.

.github/workflows/docker-build.yml

@@ -64,3 +64,27 @@ jobs:
           cache-to: type=gha,mode=max
           platforms: linux/amd64,linux/arm64
 
+      # Push to AWS ECR for App Runner auto-deploy
+      - name: Configure AWS credentials
+        if: github.event_name != 'pull_request' && github.ref == 'refs/heads/main'
+        uses: aws-actions/configure-aws-credentials@v4
+        with:
+          aws-access-key-id: ${{ secrets.AWS_ACCESS_KEY_ID }}
+          aws-secret-access-key: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
+          aws-region: ap-northeast-1
+
+      - name: Login to Amazon ECR
+        if: github.event_name != 'pull_request' && github.ref == 'refs/heads/main'
+        id: login-ecr
+        uses: aws-actions/amazon-ecr-login@v2
+
+      - name: Push to ECR (triggers App Runner auto-deploy)
+        if: github.event_name != 'pull_request' && github.ref == 'refs/heads/main'
+        env:
+          REPO_LOWER: ${{ github.repository }}
+        run: |
+          REPO_LOWER=$(echo "$REPO_LOWER" | tr '[:upper:]' '[:lower:]')
+          docker pull ghcr.io/${REPO_LOWER}:latest
+          docker tag ghcr.io/${REPO_LOWER}:latest ${{ secrets.AWS_ACCOUNT_ID }}.dkr.ecr.ap-northeast-1.amazonaws.com/next-ai-draw-io:latest
+          docker push ${{ secrets.AWS_ACCOUNT_ID }}.dkr.ecr.ap-northeast-1.amazonaws.com/next-ai-draw-io:latest
+

.gitignore

@@ -2,6 +2,8 @@
 
 # dependencies
 /node_modules
+packages/*/node_modules
+packages/*/dist
 /.pnp
 .pnp.*
 .yarn/*
@@ -40,5 +42,11 @@ yarn-error.log*
 *.tsbuildinfo
 next-env.d.ts
 push-via-ec2.sh
-.claude/settings.local.json
-.playwright-mcp/
+.claude/
+.playwright-mcp/
+# Cloudflare
+.dev.vars
+.open-next/
+.wrangler/
+.env*.local
+

Dockerfile

@@ -22,6 +22,10 @@ COPY . .
 # Disable Next.js telemetry during build
 ENV NEXT_TELEMETRY_DISABLED=1
 
+# Build-time argument for self-hosted draw.io URL
+ARG NEXT_PUBLIC_DRAWIO_BASE_URL=https://embed.diagrams.net
+ENV NEXT_PUBLIC_DRAWIO_BASE_URL=${NEXT_PUBLIC_DRAWIO_BASE_URL}
+
 # Build Next.js application (standalone mode)
 RUN npm run build
 
@@ -50,6 +54,6 @@ EXPOSE 3000
 ENV PORT=3000
 ENV HOSTNAME="0.0.0.0"
 
-# Start the application
-CMD ["node", "server.js"]
+# Start the application (HOSTNAME override needed for AWS App Runner)
+CMD ["sh", "-c", "HOSTNAME=0.0.0.0 exec node server.js"]
 

README.md

@@ -4,31 +4,45 @@
 
 **AI-Powered Diagram Creation Tool - Chat, Draw, Visualize**
 
-English | [中文](./README_CN.md) | [日本語](./README_JA.md)
+English | [中文](./docs/README_CN.md) | [日本語](./docs/README_JA.md)
 
-[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
-[![Next.js](https://img.shields.io/badge/Next.js-15.x-black)](https://nextjs.org/)
-[![TypeScript](https://img.shields.io/badge/TypeScript-5.x-blue)](https://www.typescriptlang.org/)
+[![TrendShift](https://trendshift.io/api/badge/repositories/15449)](https://next-ai-drawio.jiang.jp/)
+
+[![License: Apache 2.0](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
+[![Next.js](https://img.shields.io/badge/Next.js-16.x-black)](https://nextjs.org/)
+[![React](https://img.shields.io/badge/React-19.x-61dafb)](https://react.dev/)
 [![Sponsor](https://img.shields.io/badge/Sponsor-❤-ea4aaa)](https://github.com/sponsors/DayuanJiang)
 
-[🚀 Live Demo](https://next-ai-drawio.jiang.jp/)
+[![Live Demo](./public/live-demo-button.svg)](https://next-ai-drawio.jiang.jp/)
 
 </div>
 
 A Next.js web application that integrates AI capabilities with draw.io diagrams. Create, modify, and enhance diagrams through natural language commands and AI-assisted visualization.
 
-https://github.com/user-attachments/assets/b2eef5f3-b335-4e71-a755-dc2e80931979
 
-## Features
 
--   **LLM-Powered Diagram Creation**: Leverage Large Language Models to create and manipulate draw.io diagrams directly through natural language commands
--   **Image-Based Diagram Replication**: Upload existing diagrams or images and have the AI replicate and enhance them automatically
--   **Diagram History**: Comprehensive version control that tracks all changes, allowing you to view and restore previous versions of your diagrams before the AI editing.
--   **Interactive Chat Interface**: Communicate with AI to refine your diagrams in real-time
--   **AWS Architecture Diagram Support**: Specialized support for generating AWS architecture diagrams
--   **Animated Connectors**: Create dynamic and animated connectors between diagram elements for better visualization
+https://github.com/user-attachments/assets/9d60a3e8-4a1c-4b5e-acbb-26af2d3eabd1
 
-## **Examples**
+
+
+## Table of Contents
+- [Next AI Draw.io ](#next-ai-drawio-)
+  - [Table of Contents](#table-of-contents)
+  - [Examples](#examples)
+  - [Features](#features)
+  - [MCP Server (Preview)](#mcp-server-preview)
+  - [Getting Started](#getting-started)
+    - [Try it Online](#try-it-online)
+    - [Run with Docker (Recommended)](#run-with-docker-recommended)
+    - [Installation](#installation)
+  - [Deployment](#deployment)
+  - [Multi-Provider Support](#multi-provider-support)
+  - [How It Works](#how-it-works)
+  - [Project Structure](#project-structure)
+  - [Support \& Contact](#support--contact)
+  - [Star History](#star-history)
+
+## Examples
 
 Here are some example prompts and their generated diagrams:
 
@@ -68,37 +82,59 @@ Here are some example prompts and their generated diagrams:
 </table>
 </div>
 
-## How It Works
+## Features
 
-The application uses the following technologies:
+-   **LLM-Powered Diagram Creation**: Leverage Large Language Models to create and manipulate draw.io diagrams directly through natural language commands
+-   **Image-Based Diagram Replication**: Upload existing diagrams or images and have the AI replicate and enhance them automatically
+-   **PDF & Text File Upload**: Upload PDF documents and text files to extract content and generate diagrams from existing documents
+-   **AI Reasoning Display**: View the AI's thinking process for supported models (OpenAI o1/o3, Gemini, Claude, etc.)
+-   **Diagram History**: Comprehensive version control that tracks all changes, allowing you to view and restore previous versions of your diagrams before the AI editing.
+-   **Interactive Chat Interface**: Communicate with AI to refine your diagrams in real-time
+-   **Cloud Architecture Diagram Support**: Specialized support for generating cloud architecture diagrams (AWS, GCP, Azure)
+-   **Animated Connectors**: Create dynamic and animated connectors between diagram elements for better visualization
 
--   **Next.js**: For the frontend framework and routing
--   **Vercel AI SDK** (`ai` + `@ai-sdk/*`): For streaming AI responses and multi-provider support
--   **react-drawio**: For diagram representation and manipulation
+## MCP Server (Preview)
 
-Diagrams are represented as XML that can be rendered in draw.io. The AI processes your commands and generates or modifies this XML accordingly.
+> **Preview Feature**: This feature is experimental and may not stable.
 
-## Multi-Provider Support
+Use Next AI Draw.io with AI agents like Claude Desktop, Cursor, and VS Code via MCP (Model Context Protocol).
 
--   AWS Bedrock (default)
--   OpenAI
--   Anthropic
--   Google AI
--   Azure OpenAI
--   Ollama
--   OpenRouter
--   DeepSeek
+```json
+{
+  "mcpServers": {
+    "drawio": {
+      "command": "npx",
+      "args": ["@next-ai-drawio/mcp-server@latest"]
+    }
+  }
+}
+```
 
-All providers except AWS Bedrock and OpenRouter support custom endpoints.
+### Claude Code CLI
 
-📖 **[Detailed Provider Configuration Guide](./docs/ai-providers.md)** - See setup instructions for each provider.
+```bash
+claude mcp add drawio -- npx @next-ai-drawio/mcp-server@latest
+```
 
-**Model Requirements**: This task requires strong model capabilities for generating long-form text with strict formatting constraints (draw.io XML). Recommended models include Claude Sonnet 4.5, GPT-4o, Gemini 2.0, and DeepSeek V3/R1.
+Then ask Claude to create diagrams:
+> "Create a flowchart showing user authentication with login, MFA, and session management"
 
-Note that `claude-sonnet-4-5` has trained on draw.io diagrams with AWS logos, so if you want to create AWS architecture diagrams, this is the best choice.
+The diagram appears in your browser in real-time!
+
+See the [MCP Server README](./packages/mcp-server/README.md) for VS Code, Cursor, and other client configurations.
 
 ## Getting Started
 
+### Try it Online
+
+No installation needed! Try the app directly on our demo site:
+
+[![Live Demo](./public/live-demo-button.svg)](https://next-ai-drawio.jiang.jp/)
+
+> Note: Due to high traffic, the demo site currently uses minimax-m2. For best results, we recommend self-hosting with Claude Sonnet 4.5 or Claude Opus 4.5.
+
+> **Bring Your Own API Key**: You can use your own API key to bypass usage limits on the demo site. Click the Settings icon in the chat panel to configure your provider and API key. Your key is stored locally in your browser and is never stored on the server.
+
 ### Run with Docker (Recommended)
 
 If you just want to run it locally, the best way is to use Docker.
@@ -115,10 +151,20 @@ docker run -d -p 3000:3000 \
   ghcr.io/dayuanjiang/next-ai-draw-io:latest
 ```
 
+Or use an env file:
+
+```bash
+cp env.example .env
+# Edit .env with your configuration
+docker run -d -p 3000:3000 --env-file .env ghcr.io/dayuanjiang/next-ai-draw-io:latest
+```
+
 Open [http://localhost:3000](http://localhost:3000) in your browser.
 
 Replace the environment variables with your preferred AI provider configuration. See [Multi-Provider Support](#multi-provider-support) for available options.
 
+> **Offline Deployment:** If `embed.diagrams.net` is blocked, see [Offline Deployment](./docs/offline-deployment.md) for configuration options.
+
 ### Installation
 
 1. Clone the repository:
@@ -132,8 +178,6 @@ cd next-ai-draw-io
 
 ```bash
 npm install
-# or
-yarn install
 ```
 
 3. Configure your AI provider:
@@ -146,9 +190,10 @@ cp env.example .env.local
 
 Edit `.env.local` and configure your chosen provider:
 
--   Set `AI_PROVIDER` to your chosen provider (bedrock, openai, anthropic, google, azure, ollama, openrouter, deepseek)
+-   Set `AI_PROVIDER` to your chosen provider (bedrock, openai, anthropic, google, azure, ollama, openrouter, deepseek, siliconflow)
 -   Set `AI_MODEL` to the specific model you want to use
 -   Add the required API keys for your provider
+-   `TEMPERATURE`: Optional temperature setting (e.g., `0` for deterministic output). Leave unset for models that don't support it (e.g., reasoning models).
 -   `ACCESS_CODE_LIST`: Optional access password(s), can be comma-separated for multiple passwords.
 
 > Warning: If you do not set `ACCESS_CODE_LIST`, anyone can access your deployed site directly, which may lead to rapid depletion of your token. It is recommended to set this option.
@@ -174,6 +219,38 @@ Or you can deploy by this button.
 
 Be sure to **set the environment variables** in the Vercel dashboard as you did in your local `.env.local` file.
 
+
+## Multi-Provider Support
+
+-   AWS Bedrock (default)
+-   OpenAI
+-   Anthropic
+-   Google AI
+-   Azure OpenAI
+-   Ollama
+-   OpenRouter
+-   DeepSeek
+-   SiliconFlow
+
+All providers except AWS Bedrock and OpenRouter support custom endpoints.
+
+📖 **[Detailed Provider Configuration Guide](./docs/ai-providers.md)** - See setup instructions for each provider.
+
+**Model Requirements**: This task requires strong model capabilities for generating long-form text with strict formatting constraints (draw.io XML). Recommended models include Claude Sonnet 4.5, GPT-5.1, Gemini 3 Pro, and DeepSeek V3.2/R1.
+
+Note that `claude` series has trained on draw.io diagrams with cloud architecture logos like AWS, Azure, GCP. So if you want to create cloud architecture diagrams, this is the best choice.
+
+
+## How It Works
+
+The application uses the following technologies:
+
+-   **Next.js**: For the frontend framework and routing
+-   **Vercel AI SDK** (`ai` + `@ai-sdk/*`): For streaming AI responses and multi-provider support
+-   **react-drawio**: For diagram representation and manipulation
+
+Diagrams are represented as XML that can be rendered in draw.io. The AI processes your commands and generates or modifies this XML accordingly.
+
 ## Project Structure
 
 ```
@@ -193,14 +270,6 @@ lib/                  # Utility functions and helpers
 public/               # Static assets including example images
 ```
 
-## TODOs
-
--   [x] Allow the LLM to modify the XML instead of generating it from scratch everytime.
--   [x] Improve the smoothness of shape streaming updates.
--   [x] Add multiple AI provider support (OpenAI, Anthropic, Google, Azure, Ollama)
--   [x] Solve the bug that generation will fail for session that longer than 60s.
--   [ ] Add API config on the UI.
-
 ## Support & Contact
 
 If you find this project useful, please consider [sponsoring](https://github.com/sponsors/DayuanJiang) to help me host the live demo site!

amplify.yml

@@ -1,22 +0,0 @@
-version: 1
-frontend:
-  phases:
-    preBuild:
-      commands:
-        - npm ci --cache .npm --prefer-offline
-    build:
-      commands:
-        # Write env vars to .env.production for Next.js SSR runtime
-        - env | grep -e AI_MODEL >> .env.production
-        - env | grep -e AI_PROVIDER >> .env.production
-        - env | grep -e OPENAI_API_KEY >> .env.production
-        - env | grep -e NEXT_PUBLIC_ >> .env.production
-        - npm run build
-  artifacts:
-    baseDirectory: .next
-    files:
-      - '**/*'
-  cache:
-    paths:
-      - .next/cache/**/*
-      - .npm/**/*

app/about/cn/page.tsx

@@ -10,7 +10,18 @@ export const metadata: Metadata = {
     keywords: ["AI图表", "draw.io", "AWS架构", "GCP图表", "Azure图表", "LLM"],
 }
 
+function formatNumber(num: number): string {
+    if (num >= 1000) {
+        return `${num / 1000}k`
+    }
+    return num.toString()
+}
+
 export default function AboutCN() {
+    const dailyRequestLimit = Number(process.env.DAILY_REQUEST_LIMIT) || 20
+    const dailyTokenLimit = Number(process.env.DAILY_TOKEN_LIMIT) || 500000
+    const tpmLimit = Number(process.env.TPM_LIMIT) || 50000
+
     return (
         <div className="min-h-screen bg-gray-50">
             {/* Navigation */}
@@ -85,12 +96,124 @@ export default function AboutCN() {
                         </div>
                     </div>
 
-                    <div className="bg-amber-50 border border-amber-200 rounded-lg p-4 mb-6">
-                        <p className="text-amber-800">
-                            本应用设计运行于 Claude Opus 4.5
-                            以获得最佳性能。但由于流量超出预期，运行顶级模型的成本变得难以承受。为避免服务中断并控制成本，我已将后端切换至
-                            Claude Haiku 4.5。
-                        </p>
+                    <div className="relative mb-8 rounded-2xl bg-gradient-to-br from-amber-50 via-orange-50 to-rose-50 p-[1px] shadow-lg">
+                        <div className="absolute inset-0 rounded-2xl bg-gradient-to-br from-amber-400 via-orange-400 to-rose-400 opacity-20" />
+                        <div className="relative rounded-2xl bg-white/80 backdrop-blur-sm p-6">
+                            {/* Header */}
+                            <div className="mb-4">
+                                <h3 className="text-lg font-bold text-gray-900 tracking-tight">
+                                    模型变更与用量限制{" "}
+                                    <span className="text-sm text-amber-600 font-medium italic font-normal">
+                                        (或者说：我的钱包顶不住了)
+                                    </span>
+                                </h3>
+                            </div>
+
+                            {/* Story */}
+                            <div className="space-y-3 text-sm text-gray-700 leading-relaxed mb-5">
+                                <p>
+                                    大家对这个项目的热情太高了——看来大家都真的很喜欢画图！但这也带来了一个幸福的烦恼：我们经常触发出上游
+                                    AI 接口的频率限制
+                                    (TPS/TPM)。一旦超限，系统就会暂停，导致请求失败。
+                                </p>
+                                <p>
+                                    由于使用量过高，我已将模型从 Claude 更换为{" "}
+                                    <span className="font-semibold text-amber-700">
+                                        minimax-m2
+                                    </span>
+                                    ，以降低成本。
+                                </p>
+                                <p>
+                                    作为一个
+                                    <span className="font-semibold text-amber-700">
+                                        独立开发者
+                                    </span>
+                                    ，目前的 API
+                                    费用全是我自己在掏腰包（纯属为爱发电）。为了保证服务能细水长流，同时也为了避免我个人陷入财务危机，我还设置了以下临时用量限制：
+                                </p>
+                            </div>
+
+                            {/* Limits Cards */}
+                            <div className="grid grid-cols-2 gap-3 mb-5">
+                                <div className="rounded-xl bg-gradient-to-br from-amber-100 to-orange-100 p-4 text-center">
+                                    <div className="text-xs font-medium text-amber-700 uppercase tracking-wide mb-1">
+                                        Token 用量
+                                    </div>
+                                    <div className="text-lg font-bold text-gray-900">
+                                        {formatNumber(tpmLimit)}
+                                        <span className="text-sm font-normal text-gray-600">
+                                            /分钟
+                                        </span>
+                                    </div>
+                                    <div className="text-lg font-bold text-gray-900">
+                                        {formatNumber(dailyTokenLimit)}
+                                        <span className="text-sm font-normal text-gray-600">
+                                            /天
+                                        </span>
+                                    </div>
+                                </div>
+                                <div className="rounded-xl bg-gradient-to-br from-amber-100 to-orange-100 p-4 text-center">
+                                    <div className="text-xs font-medium text-amber-700 uppercase tracking-wide mb-1">
+                                        每日请求数
+                                    </div>
+                                    <div className="text-2xl font-bold text-gray-900">
+                                        {dailyRequestLimit}
+                                    </div>
+                                    <div className="text-sm text-gray-600">
+                                        次
+                                    </div>
+                                </div>
+                            </div>
+
+                            {/* Divider */}
+                            <div className="flex items-center gap-3 my-5">
+                                <div className="flex-1 h-px bg-gradient-to-r from-transparent via-amber-300 to-transparent" />
+                            </div>
+
+                            {/* Bring Your Own Key */}
+                            <div className="text-center mb-5">
+                                <h4 className="text-base font-bold text-gray-900 mb-2">
+                                    使用自己的 API Key
+                                </h4>
+                                <p className="text-sm text-gray-600 mb-2 max-w-md mx-auto">
+                                    您可以使用自己的 API Key
+                                    来绕过这些限制。点击聊天面板中的设置图标即可配置您的
+                                    Provider 和 API Key。
+                                </p>
+                                <p className="text-xs text-gray-500 max-w-md mx-auto">
+                                    您的 Key
+                                    仅保存在浏览器本地，不会被存储在服务器上。
+                                </p>
+                            </div>
+
+                            {/* Divider */}
+                            <div className="flex items-center gap-3 mb-5">
+                                <div className="flex-1 h-px bg-gradient-to-r from-transparent via-amber-300 to-transparent" />
+                            </div>
+
+                            {/* Sponsorship CTA */}
+                            <div className="text-center">
+                                <h4 className="text-base font-bold text-gray-900 mb-2">
+                                    寻求赞助 (求大佬捞一把)
+                                </h4>
+                                <p className="text-sm text-gray-600 mb-4 max-w-md mx-auto">
+                                    要想彻底解除这些限制，扩容后端是唯一的办法。我正在积极寻求
+                                    AI API 提供商或云平台的赞助。
+                                </p>
+                                <p className="text-sm text-gray-600 mb-4 max-w-md mx-auto">
+                                    作为回报（无论是额度支持还是资金支持），我将在
+                                    GitHub 仓库和 Live Demo
+                                    网站的显眼位置展示贵公司的 Logo
+                                    作为平台赞助商。
+                                </p>
+                                <a
+                                    href="mailto:me@jiang.jp"
+                                    className="inline-flex items-center gap-2 px-5 py-2.5 rounded-full bg-gradient-to-r from-amber-500 to-orange-500 text-white font-medium text-sm shadow-md hover:shadow-lg hover:scale-105 transition-all duration-200"
+                                >
+                                    联系我
+                                </a>
+                            </div>
+                        </div>
                     </div>
 
                     <p className="text-gray-700">

app/about/ja/page.tsx

@@ -17,7 +17,18 @@ export const metadata: Metadata = {
     ],
 }
 
+function formatNumber(num: number): string {
+    if (num >= 1000) {
+        return `${num / 1000}k`
+    }
+    return num.toString()
+}
+
 export default function AboutJA() {
+    const dailyRequestLimit = Number(process.env.DAILY_REQUEST_LIMIT) || 20
+    const dailyTokenLimit = Number(process.env.DAILY_TOKEN_LIMIT) || 500000
+    const tpmLimit = Number(process.env.TPM_LIMIT) || 50000
+
     return (
         <div className="min-h-screen bg-gray-50">
             {/* Navigation */}
@@ -93,13 +104,121 @@ export default function AboutJA() {
                         </div>
                     </div>
 
-                    <div className="bg-amber-50 border border-amber-200 rounded-lg p-4 mb-6">
-                        <p className="text-amber-800">
-                            本アプリは最高のパフォーマンスを発揮するため、Claude
-                            Opus 4.5
-                            で動作するよう設計されています。しかし、予想以上のトラフィックにより、最上位モデルの運用コストが負担となっています。サービスの中断を避け、コストを管理するため、バックエンドを
-                            Claude Haiku 4.5 に切り替えました。
-                        </p>
+                    <div className="relative mb-8 rounded-2xl bg-gradient-to-br from-amber-50 via-orange-50 to-rose-50 p-[1px] shadow-lg">
+                        <div className="absolute inset-0 rounded-2xl bg-gradient-to-br from-amber-400 via-orange-400 to-rose-400 opacity-20" />
+                        <div className="relative rounded-2xl bg-white/80 backdrop-blur-sm p-6">
+                            {/* Header */}
+                            <div className="mb-4">
+                                <h3 className="text-lg font-bold text-gray-900 tracking-tight">
+                                    モデル変更と利用制限について{" "}
+                                    <span className="text-sm text-amber-600 font-medium italic font-normal">
+                                        （別名：お財布が悲鳴を上げています）
+                                    </span>
+                                </h3>
+                            </div>
+
+                            {/* Story */}
+                            <div className="space-y-3 text-sm text-gray-700 leading-relaxed mb-5">
+                                <p>
+                                    予想以上の反響をいただき、ありがとうございます！皆様にダイアグラム作成を楽しんでいただいているのは嬉しい限りですが、その熱量により
+                                    AI API のレート制限 (TPS/TPM)
+                                    に頻繁に引っかかってしまっています。制限に達するとシステムが一時停止し、エラーが発生してしまいます。
+                                </p>
+                                <p>
+                                    利用量の増加に伴い、コスト削減のためモデルを
+                                    Claude から{" "}
+                                    <span className="font-semibold text-amber-700">
+                                        minimax-m2
+                                    </span>{" "}
+                                    に変更しました。
+                                </p>
+                                <p>
+                                    私は現在、
+                                    <span className="font-semibold text-amber-700">
+                                        個人開発者
+                                    </span>
+                                    として API
+                                    費用を全額自腹で負担しています。サービスを継続し、かつ私自身が借金を背負わないようにするため（笑）、一時的に以下の利用制限も設けさせていただきました。
+                                </p>
+                            </div>
+
+                            {/* Limits Cards */}
+                            <div className="grid grid-cols-2 gap-3 mb-5">
+                                <div className="rounded-xl bg-gradient-to-br from-amber-100 to-orange-100 p-4 text-center">
+                                    <div className="text-xs font-medium text-amber-700 uppercase tracking-wide mb-1">
+                                        トークン使用量
+                                    </div>
+                                    <div className="text-lg font-bold text-gray-900">
+                                        {formatNumber(tpmLimit)}
+                                        <span className="text-sm font-normal text-gray-600">
+                                            /分
+                                        </span>
+                                    </div>
+                                    <div className="text-lg font-bold text-gray-900">
+                                        {formatNumber(dailyTokenLimit)}
+                                        <span className="text-sm font-normal text-gray-600">
+                                            /日
+                                        </span>
+                                    </div>
+                                </div>
+                                <div className="rounded-xl bg-gradient-to-br from-amber-100 to-orange-100 p-4 text-center">
+                                    <div className="text-xs font-medium text-amber-700 uppercase tracking-wide mb-1">
+                                        1日のリクエスト数
+                                    </div>
+                                    <div className="text-2xl font-bold text-gray-900">
+                                        {dailyRequestLimit}
+                                    </div>
+                                    <div className="text-sm text-gray-600">
+                                        回
+                                    </div>
+                                </div>
+                            </div>
+
+                            {/* Divider */}
+                            <div className="flex items-center gap-3 my-5">
+                                <div className="flex-1 h-px bg-gradient-to-r from-transparent via-amber-300 to-transparent" />
+                            </div>
+
+                            {/* Bring Your Own Key */}
+                            <div className="text-center mb-5">
+                                <h4 className="text-base font-bold text-gray-900 mb-2">
+                                    自分のAPIキーを使用
+                                </h4>
+                                <p className="text-sm text-gray-600 mb-2 max-w-md mx-auto">
+                                    自分のAPIキーを使用することで、これらの制限を回避できます。チャットパネルの設定アイコンをクリックして、プロバイダーとAPIキーを設定してください。
+                                </p>
+                                <p className="text-xs text-gray-500 max-w-md mx-auto">
+                                    キーはブラウザのローカルに保存され、サーバーには保存されません。
+                                </p>
+                            </div>
+
+                            {/* Divider */}
+                            <div className="flex items-center gap-3 mb-5">
+                                <div className="flex-1 h-px bg-gradient-to-r from-transparent via-amber-300 to-transparent" />
+                            </div>
+
+                            {/* Sponsorship CTA */}
+                            <div className="text-center">
+                                <h4 className="text-base font-bold text-gray-900 mb-2">
+                                    スポンサー募集
+                                </h4>
+                                <p className="text-sm text-gray-600 mb-4 max-w-md mx-auto">
+                                    これらの制限を取り払い、バックエンドをスケールさせるには皆様の支援が必要です。現在、AI
+                                    API
+                                    プロバイダー様やクラウドプラットフォーム様からのスポンサー支援を積極的に募集しています。
+                                </p>
+                                <p className="text-sm text-gray-600 mb-4 max-w-md mx-auto">
+                                    ご支援（クレジット提供や資金援助）をいただける場合、GitHub
+                                    リポジトリおよびデモサイトにて、プラットフォームスポンサーとして貴社を大々的にご紹介させていただきます。
+                                </p>
+                                <a
+                                    href="mailto:me@jiang.jp"
+                                    className="inline-flex items-center gap-2 px-5 py-2.5 rounded-full bg-gradient-to-r from-amber-500 to-orange-500 text-white font-medium text-sm shadow-md hover:shadow-lg hover:scale-105 transition-all duration-200"
+                                >
+                                    お問い合わせ
+                                </a>
+                            </div>
+                        </div>
                     </div>
 
                     <p className="text-gray-700">

app/about/page.tsx

@@ -17,7 +17,18 @@ export const metadata: Metadata = {
     ],
 }
 
+function formatNumber(num: number): string {
+    if (num >= 1000) {
+        return `${num / 1000}k`
+    }
+    return num.toString()
+}
+
 export default function About() {
+    const dailyRequestLimit = Number(process.env.DAILY_REQUEST_LIMIT) || 20
+    const dailyTokenLimit = Number(process.env.DAILY_TOKEN_LIMIT) || 500000
+    const tpmLimit = Number(process.env.TPM_LIMIT) || 50000
+
     return (
         <div className="min-h-screen bg-gray-50">
             {/* Navigation */}
@@ -93,15 +104,134 @@ export default function About() {
                         </div>
                     </div>
 
-                    <div className="bg-amber-50 border border-amber-200 rounded-lg p-4 mb-6">
-                        <p className="text-amber-800">
-                            This app is designed to run on Claude Opus 4.5 for
-                            best performance. However, due to
-                            higher-than-expected traffic, running the top-tier
-                            model has become cost-prohibitive. To avoid service
-                            interruptions and manage costs, I have switched the
-                            backend to Claude Haiku 4.5.
-                        </p>
+                    <div className="relative mb-8 rounded-2xl bg-gradient-to-br from-amber-50 via-orange-50 to-rose-50 p-[1px] shadow-lg">
+                        <div className="absolute inset-0 rounded-2xl bg-gradient-to-br from-amber-400 via-orange-400 to-rose-400 opacity-20" />
+                        <div className="relative rounded-2xl bg-white/80 backdrop-blur-sm p-6">
+                            {/* Header */}
+                            <div className="mb-4">
+                                <h3 className="text-lg font-bold text-gray-900 tracking-tight">
+                                    Model Change & Usage Limits{" "}
+                                    <span className="text-sm text-amber-600 font-medium italic font-normal">
+                                        (Or: Why My Wallet is Crying)
+                                    </span>
+                                </h3>
+                            </div>
+
+                            {/* Story */}
+                            <div className="space-y-3 text-sm text-gray-700 leading-relaxed mb-5">
+                                <p>
+                                    The response to this project has been
+                                    incredible—you all love making diagrams!
+                                    However, this enthusiasm means we are
+                                    frequently hitting the AI API rate limits
+                                    (TPS/TPM). When this happens, the system
+                                    pauses, leading to failed requests.
+                                </p>
+                                <p>
+                                    Due to the high usage, I have changed the
+                                    model from Claude to{" "}
+                                    <span className="font-semibold text-amber-700">
+                                        minimax-m2
+                                    </span>
+                                    , which is more cost-effective.
+                                </p>
+                                <p>
+                                    As an{" "}
+                                    <span className="font-semibold text-amber-700">
+                                        indie developer
+                                    </span>
+                                    , I am currently footing the entire API
+                                    bill. To keep the lights on and ensure the
+                                    service remains available to everyone
+                                    without sending me into debt, I have also
+                                    implemented the following temporary caps:
+                                </p>
+                            </div>
+
+                            {/* Limits Cards */}
+                            <div className="grid grid-cols-2 gap-3 mb-5">
+                                <div className="rounded-xl bg-gradient-to-br from-amber-100 to-orange-100 p-4 text-center">
+                                    <div className="text-xs font-medium text-amber-700 uppercase tracking-wide mb-1">
+                                        Token Usage
+                                    </div>
+                                    <div className="text-lg font-bold text-gray-900">
+                                        {formatNumber(tpmLimit)}
+                                        <span className="text-sm font-normal text-gray-600">
+                                            /min
+                                        </span>
+                                    </div>
+                                    <div className="text-lg font-bold text-gray-900">
+                                        {formatNumber(dailyTokenLimit)}
+                                        <span className="text-sm font-normal text-gray-600">
+                                            /day
+                                        </span>
+                                    </div>
+                                </div>
+                                <div className="rounded-xl bg-gradient-to-br from-amber-100 to-orange-100 p-4 text-center">
+                                    <div className="text-xs font-medium text-amber-700 uppercase tracking-wide mb-1">
+                                        Daily Requests
+                                    </div>
+                                    <div className="text-2xl font-bold text-gray-900">
+                                        {dailyRequestLimit}
+                                    </div>
+                                    <div className="text-sm text-gray-600">
+                                        requests
+                                    </div>
+                                </div>
+                            </div>
+
+                            {/* Divider */}
+                            <div className="flex items-center gap-3 my-5">
+                                <div className="flex-1 h-px bg-gradient-to-r from-transparent via-amber-300 to-transparent" />
+                            </div>
+
+                            {/* Bring Your Own Key */}
+                            <div className="text-center mb-5">
+                                <h4 className="text-base font-bold text-gray-900 mb-2">
+                                    Bring Your Own API Key
+                                </h4>
+                                <p className="text-sm text-gray-600 mb-2 max-w-md mx-auto">
+                                    You can use your own API key to bypass these
+                                    limits. Click the Settings icon in the chat
+                                    panel to configure your provider and API
+                                    key.
+                                </p>
+                                <p className="text-xs text-gray-500 max-w-md mx-auto">
+                                    Your key is stored locally in your browser
+                                    and is never stored on the server.
+                                </p>
+                            </div>
+
+                            {/* Divider */}
+                            <div className="flex items-center gap-3 mb-5">
+                                <div className="flex-1 h-px bg-gradient-to-r from-transparent via-amber-300 to-transparent" />
+                            </div>
+
+                            {/* Sponsorship CTA */}
+                            <div className="text-center">
+                                <h4 className="text-base font-bold text-gray-900 mb-2">
+                                    Call for Sponsorship
+                                </h4>
+                                <p className="text-sm text-gray-600 mb-4 max-w-md mx-auto">
+                                    Scaling the backend is the only way to
+                                    remove these limits. I am actively seeking
+                                    sponsorship from AI API providers or Cloud
+                                    Platforms.
+                                </p>
+                                <p className="text-sm text-gray-600 mb-4 max-w-md mx-auto">
+                                    In return for support (credits or funding),
+                                    I will prominently feature your company as a
+                                    platform sponsor on both the GitHub
+                                    repository and the live demo site.
+                                </p>
+                                <a
+                                    href="mailto:me@jiang.jp"
+                                    className="inline-flex items-center gap-2 px-5 py-2.5 rounded-full bg-gradient-to-r from-amber-500 to-orange-500 text-white font-medium text-sm shadow-md hover:shadow-lg hover:scale-105 transition-all duration-200"
+                                >
+                                    Contact Me
+                                </a>
+                            </div>
+                        </div>
                     </div>
 
                     <p className="text-gray-700">

app/api/chat/route.ts

@@ -1,11 +1,16 @@
 import {
+    APICallError,
     convertToModelMessages,
     createUIMessageStream,
     createUIMessageStreamResponse,
+    InvalidToolInputError,
+    LoadAPIKeyError,
+    stepCountIs,
     streamText,
 } from "ai"
+import { jsonrepair } from "jsonrepair"
 import { z } from "zod"
-import { getAIModel } from "@/lib/ai-providers"
+import { getAIModel, supportsPromptCaching } from "@/lib/ai-providers"
 import { findCachedResponse } from "@/lib/cached-responses"
 import {
     getTelemetryConfig,
@@ -15,7 +20,7 @@ import {
 } from "@/lib/langfuse"
 import { getSystemPrompt } from "@/lib/system-prompts"
 
-export const maxDuration = 300
+export const maxDuration = 120
 
 // File upload limits (must match client-side)
 const MAX_FILE_SIZE = 2 * 1024 * 1024 // 2MB
@@ -40,7 +45,7 @@ function validateFileParts(messages: any[]): {
     for (const filePart of fileParts) {
         // Data URLs format: data:image/png;base64,<data>
         // Base64 increases size by ~33%, so we check the decoded size
-        if (filePart.url && filePart.url.startsWith("data:")) {
+        if (filePart.url?.startsWith("data:")) {
             const base64Data = filePart.url.split(",")[1]
             if (base64Data) {
                 const sizeInBytes = Math.ceil((base64Data.length * 3) / 4)
@@ -63,6 +68,47 @@ function isMinimalDiagram(xml: string): boolean {
     return !stripped.includes('id="2"')
 }
 
+// Helper function to replace historical tool call XML with placeholders
+// This reduces token usage and forces LLM to rely on the current diagram XML (source of truth)
+// Also fixes invalid/undefined inputs from interrupted streaming
+function replaceHistoricalToolInputs(messages: any[]): any[] {
+    return messages.map((msg) => {
+        if (msg.role !== "assistant" || !Array.isArray(msg.content)) {
+            return msg
+        }
+        const replacedContent = msg.content
+            .map((part: any) => {
+                if (part.type === "tool-call") {
+                    const toolName = part.toolName
+                    // Fix invalid/undefined inputs from interrupted streaming
+                    if (
+                        !part.input ||
+                        typeof part.input !== "object" ||
+                        Object.keys(part.input).length === 0
+                    ) {
+                        // Skip tool calls with invalid inputs entirely
+                        return null
+                    }
+                    if (
+                        toolName === "display_diagram" ||
+                        toolName === "edit_diagram"
+                    ) {
+                        return {
+                            ...part,
+                            input: {
+                                placeholder:
+                                    "[XML content replaced - see current diagram XML in system context]",
+                            },
+                        }
+                    }
+                }
+                return part
+            })
+            .filter(Boolean) // Remove null entries (invalid tool calls)
+        return { ...msg, content: replacedContent }
+    })
+}
+
 // Helper function to create cached stream response
 function createCachedStreamResponse(xml: string): Response {
     const toolCallId = `cached-${Date.now()}`
@@ -112,7 +158,7 @@ async function handleChatRequest(req: Request): Promise<Response> {
         }
     }
 
-    const { messages, xml, sessionId } = await req.json()
+    const { messages, xml, previousXml, sessionId } = await req.json()
 
     // Get user IP for Langfuse tracking
     const forwardedFor = req.headers.get("x-forwarded-for")
@@ -125,9 +171,9 @@ async function handleChatRequest(req: Request): Promise<Response> {
             : undefined
 
     // Extract user input text for Langfuse trace
-    const currentMessage = messages[messages.length - 1]
+    const lastMessage = messages[messages.length - 1]
     const userInputText =
-        currentMessage?.parts?.find((p: any) => p.type === "text")?.text || ""
+        lastMessage?.parts?.find((p: any) => p.type === "text")?.text || ""
 
     // Update Langfuse trace with input, session, and user
     setTraceInput({
@@ -155,26 +201,34 @@ async function handleChatRequest(req: Request): Promise<Response> {
         const cached = findCachedResponse(textPart?.text || "", !!filePart)
 
         if (cached) {
-            console.log(
-                "[Cache] Returning cached response for:",
-                textPart?.text,
-            )
             return createCachedStreamResponse(cached.xml)
         }
     }
     // === CACHE CHECK END ===
 
-    // Get AI model from environment configuration
-    const { model, providerOptions, headers, modelId } = getAIModel()
+    // Read client AI provider overrides from headers
+    const clientOverrides = {
+        provider: req.headers.get("x-ai-provider"),
+        baseUrl: req.headers.get("x-ai-base-url"),
+        apiKey: req.headers.get("x-ai-api-key"),
+        modelId: req.headers.get("x-ai-model"),
+    }
+
+    // Read minimal style preference from header
+    const minimalStyle = req.headers.get("x-minimal-style") === "true"
+
+    // Get AI model with optional client overrides
+    const { model, providerOptions, headers, modelId } =
+        getAIModel(clientOverrides)
+
+    // Check if model supports prompt caching
+    const shouldCache = supportsPromptCaching(modelId)
+    console.log(
+        `[Prompt Caching] ${shouldCache ? "ENABLED" : "DISABLED"} for model: ${modelId}`,
+    )
 
     // Get the appropriate system prompt based on model (extended for Opus/Haiku 4.5)
-    const systemMessage = getSystemPrompt(modelId)
-
-    const lastMessage = messages[messages.length - 1]
-
-    // Extract text from the last message parts
-    const lastMessageText =
-        lastMessage.parts?.find((part: any) => part.type === "text")?.text || ""
+    const systemMessage = getSystemPrompt(modelId, minimalStyle)
 
     // Extract file parts (images) from the last message
     const fileParts =
@@ -183,19 +237,114 @@ async function handleChatRequest(req: Request): Promise<Response> {
     // User input only - XML is now in a separate cached system message
     const formattedUserInput = `User input:
 """md
-${lastMessageText}
+${userInputText}
 """`
 
     // Convert UIMessages to ModelMessages and add system message
     const modelMessages = convertToModelMessages(messages)
 
+    // DEBUG: Log incoming messages structure
+    console.log("[route.ts] Incoming messages count:", messages.length)
+    messages.forEach((msg: any, idx: number) => {
+        console.log(
+            `[route.ts] Message ${idx} role:`,
+            msg.role,
+            "parts count:",
+            msg.parts?.length,
+        )
+        if (msg.parts) {
+            msg.parts.forEach((part: any, partIdx: number) => {
+                if (
+                    part.type === "tool-invocation" ||
+                    part.type === "tool-result"
+                ) {
+                    console.log(`[route.ts]   Part ${partIdx}:`, {
+                        type: part.type,
+                        toolName: part.toolName,
+                        hasInput: !!part.input,
+                        inputType: typeof part.input,
+                        inputKeys:
+                            part.input && typeof part.input === "object"
+                                ? Object.keys(part.input)
+                                : null,
+                    })
+                }
+            })
+        }
+    })
+
+    // Replace historical tool call XML with placeholders to reduce tokens
+    // Disabled by default - some models (e.g. minimax) copy placeholders instead of generating XML
+    const enableHistoryReplace =
+        process.env.ENABLE_HISTORY_XML_REPLACE === "true"
+    const placeholderMessages = enableHistoryReplace
+        ? replaceHistoricalToolInputs(modelMessages)
+        : modelMessages
+
     // Filter out messages with empty content arrays (Bedrock API rejects these)
     // This is a safety measure - ideally convertToModelMessages should handle all cases
-    let enhancedMessages = modelMessages.filter(
+    let enhancedMessages = placeholderMessages.filter(
         (msg: any) =>
             msg.content && Array.isArray(msg.content) && msg.content.length > 0,
     )
 
+    // Filter out tool-calls with invalid inputs (from failed repair or interrupted streaming)
+    // Bedrock API rejects messages where toolUse.input is not a valid JSON object
+    enhancedMessages = enhancedMessages
+        .map((msg: any) => {
+            if (msg.role !== "assistant" || !Array.isArray(msg.content)) {
+                return msg
+            }
+            const filteredContent = msg.content.filter((part: any) => {
+                if (part.type === "tool-call") {
+                    // Check if input is a valid object (not null, undefined, or empty)
+                    if (
+                        !part.input ||
+                        typeof part.input !== "object" ||
+                        Object.keys(part.input).length === 0
+                    ) {
+                        console.warn(
+                            `[route.ts] Filtering out tool-call with invalid input:`,
+                            { toolName: part.toolName, input: part.input },
+                        )
+                        return false
+                    }
+                }
+                return true
+            })
+            return { ...msg, content: filteredContent }
+        })
+        .filter((msg: any) => msg.content && msg.content.length > 0)
+
+    // DEBUG: Log modelMessages structure (what's being sent to AI)
+    console.log("[route.ts] Model messages count:", enhancedMessages.length)
+    enhancedMessages.forEach((msg: any, idx: number) => {
+        console.log(
+            `[route.ts] ModelMsg ${idx} role:`,
+            msg.role,
+            "content count:",
+            msg.content?.length,
+        )
+        if (msg.content) {
+            msg.content.forEach((part: any, partIdx: number) => {
+                if (part.type === "tool-call" || part.type === "tool-result") {
+                    console.log(`[route.ts]   Content ${partIdx}:`, {
+                        type: part.type,
+                        toolName: part.toolName,
+                        hasInput: !!part.input,
+                        inputType: typeof part.input,
+                        inputValue:
+                            part.input === undefined
+                                ? "undefined"
+                                : part.input === null
+                                  ? "null"
+                                  : "object",
+                    })
+                }
+            })
+        }
+    })
+
     // Update the last message with user input only (XML moved to separate cached system message)
     if (enhancedMessages.length >= 1) {
         const lastModelMessage = enhancedMessages[enhancedMessages.length - 1]
@@ -224,7 +373,7 @@ ${lastMessageText}
     // Add cache point to the last assistant message in conversation history
     // This caches the entire conversation prefix for subsequent requests
     // Strategy: system (cached) + history with last assistant (cached) + new user message
-    if (enhancedMessages.length >= 2) {
+    if (shouldCache && enhancedMessages.length >= 2) {
         // Find the last assistant message (should be second-to-last, before current user message)
         for (let i = enhancedMessages.length - 2; i >= 0; i--) {
             if (enhancedMessages[i].role === "assistant") {
@@ -249,17 +398,21 @@ ${lastMessageText}
         {
             role: "system" as const,
             content: systemMessage,
-            providerOptions: {
-                bedrock: { cachePoint: { type: "default" } },
-            },
+            ...(shouldCache && {
+                providerOptions: {
+                    bedrock: { cachePoint: { type: "default" } },
+                },
+            }),
         },
-        // Cache breakpoint 2: Current diagram XML context
+        // Cache breakpoint 2: Previous and Current diagram XML context
         {
             role: "system" as const,
-            content: `Current diagram XML:\n"""xml\n${xml || ""}\n"""\nWhen using edit_diagram, COPY search patterns exactly from this XML - attribute order matters!`,
-            providerOptions: {
-                bedrock: { cachePoint: { type: "default" } },
-            },
+            content: `${previousXml ? `Previous diagram XML (before user's last message):\n"""xml\n${previousXml}\n"""\n\n` : ""}Current diagram XML (AUTHORITATIVE - the source of truth):\n"""xml\n${xml || ""}\n"""\n\nIMPORTANT: The "Current diagram XML" is the SINGLE SOURCE OF TRUTH for what's on the canvas right now. The user can manually add, delete, or modify shapes directly in draw.io. Always count and describe elements based on the CURRENT XML, not on what you previously generated. If both previous and current XML are shown, compare them to understand what the user changed. When using edit_diagram, COPY search patterns exactly from the CURRENT XML - attribute order matters!`,
+            ...(shouldCache && {
+                providerOptions: {
+                    bedrock: { cachePoint: { type: "default" } },
+                },
+            }),
         },
     ]
 
@@ -267,8 +420,73 @@ ${lastMessageText}
 
     const result = streamText({
         model,
+        ...(process.env.MAX_OUTPUT_TOKENS && {
+            maxOutputTokens: parseInt(process.env.MAX_OUTPUT_TOKENS, 10),
+        }),
+        stopWhen: stepCountIs(5),
+        // Repair truncated tool calls when maxOutputTokens is reached mid-JSON
+        experimental_repairToolCall: async ({ toolCall, error }) => {
+            // DEBUG: Log what we're trying to repair
+            console.log(`[repairToolCall] Tool: ${toolCall.toolName}`)
+            console.log(
+                `[repairToolCall] Error: ${error.name} - ${error.message}`,
+            )
+            console.log(`[repairToolCall] Input type: ${typeof toolCall.input}`)
+            console.log(`[repairToolCall] Input value:`, toolCall.input)
+
+            // Only attempt repair for invalid tool input (broken JSON from truncation)
+            if (
+                error instanceof InvalidToolInputError ||
+                error.name === "AI_InvalidToolInputError"
+            ) {
+                try {
+                    // Pre-process to fix common LLM JSON errors that jsonrepair can't handle
+                    let inputToRepair = toolCall.input
+                    if (typeof inputToRepair === "string") {
+                        // Fix `:=` instead of `: ` (LLM sometimes generates this)
+                        inputToRepair = inputToRepair.replace(/:=/g, ": ")
+                        // Fix `= "` instead of `: "`
+                        inputToRepair = inputToRepair.replace(/=\s*"/g, ': "')
+                    }
+                    // Use jsonrepair to fix truncated JSON
+                    const repairedInput = jsonrepair(inputToRepair)
+                    console.log(
+                        `[repairToolCall] Repaired truncated JSON for tool: ${toolCall.toolName}`,
+                    )
+                    return { ...toolCall, input: repairedInput }
+                } catch (repairError) {
+                    console.warn(
+                        `[repairToolCall] Failed to repair JSON for tool: ${toolCall.toolName}`,
+                        repairError,
+                    )
+                    // Return a placeholder input to avoid API errors in multi-step
+                    // The tool will fail gracefully on client side
+                    if (toolCall.toolName === "edit_diagram") {
+                        return {
+                            ...toolCall,
+                            input: {
+                                operations: [],
+                                _error: "JSON repair failed - no operations to apply",
+                            },
+                        }
+                    }
+                    if (toolCall.toolName === "display_diagram") {
+                        return {
+                            ...toolCall,
+                            input: {
+                                xml: "",
+                                _error: "JSON repair failed - empty diagram",
+                            },
+                        }
+                    }
+                    return null
+                }
+            }
+            // Don't attempt to repair other errors (like NoSuchToolError)
+            return null
+        },
         messages: allMessages,
-        ...(providerOptions && { providerOptions }),
+        ...(providerOptions && { providerOptions }), // This now includes all reasoning configs
         ...(headers && { headers }),
         // Langfuse telemetry config (returns undefined if not configured)
         ...(getTelemetryConfig({ sessionId: validSessionId, userId }) && {
@@ -277,14 +495,8 @@ ${lastMessageText}
                 userId,
             }),
         }),
-        onFinish: ({ text, usage, providerMetadata }) => {
-            console.log(
-                "[Cache] Full providerMetadata:",
-                JSON.stringify(providerMetadata, null, 2),
-            )
-            console.log("[Cache] Usage:", JSON.stringify(usage, null, 2))
+        onFinish: ({ text, usage }) => {
             // Pass usage to Langfuse (Bedrock streaming doesn't auto-report tokens to telemetry)
-            // AI SDK uses inputTokens/outputTokens, Langfuse expects promptTokens/completionTokens
             setTraceOutput(text, {
                 promptTokens: usage?.inputTokens,
                 completionTokens: usage?.outputTokens,
@@ -293,36 +505,32 @@ ${lastMessageText}
         tools: {
             // Client-side tool that will be executed on the client
             display_diagram: {
-                description: `Display a diagram on draw.io. Pass the XML content inside <root> tags.
+                description: `Display a diagram on draw.io. Pass ONLY the mxCell elements - wrapper tags and root cells are added automatically.
 
 VALIDATION RULES (XML will be rejected if violated):
-1. All mxCell elements must be DIRECT children of <root> - never nested
-2. Every mxCell needs a unique id
-3. Every mxCell (except id="0") needs a valid parent attribute
-4. Edge source/target must reference existing cell IDs
-5. Escape special chars in values: &lt; &gt; &amp; &quot;
-6. Always start with: <mxCell id="0"/><mxCell id="1" parent="0"/>
+1. Generate ONLY mxCell elements - NO wrapper tags (<mxfile>, <mxGraphModel>, <root>)
+2. Do NOT include root cells (id="0" or id="1") - they are added automatically
+3. All mxCell elements must be siblings - never nested
+4. Every mxCell needs a unique id (start from "2")
+5. Every mxCell needs a valid parent attribute (use "1" for top-level)
+6. Escape special chars in values: &lt; &gt; &amp; &quot;
 
-Example with swimlanes and edges (note: all mxCells are siblings):
-<root>
-  <mxCell id="0"/>
-  <mxCell id="1" parent="0"/>
-  <mxCell id="lane1" value="Frontend" style="swimlane;" vertex="1" parent="1">
-    <mxGeometry x="40" y="40" width="200" height="200" as="geometry"/>
-  </mxCell>
-  <mxCell id="step1" value="Step 1" style="rounded=1;" vertex="1" parent="lane1">
-    <mxGeometry x="20" y="60" width="160" height="40" as="geometry"/>
-  </mxCell>
-  <mxCell id="lane2" value="Backend" style="swimlane;" vertex="1" parent="1">
-    <mxGeometry x="280" y="40" width="200" height="200" as="geometry"/>
-  </mxCell>
-  <mxCell id="step2" value="Step 2" style="rounded=1;" vertex="1" parent="lane2">
-    <mxGeometry x="20" y="60" width="160" height="40" as="geometry"/>
-  </mxCell>
-  <mxCell id="edge1" style="edgeStyle=orthogonalEdgeStyle;endArrow=classic;" edge="1" parent="1" source="step1" target="step2">
-    <mxGeometry relative="1" as="geometry"/>
-  </mxCell>
-</root>
+Example (generate ONLY this - no wrapper tags):
+<mxCell id="lane1" value="Frontend" style="swimlane;" vertex="1" parent="1">
+  <mxGeometry x="40" y="40" width="200" height="200" as="geometry"/>
+</mxCell>
+<mxCell id="step1" value="Step 1" style="rounded=1;" vertex="1" parent="lane1">
+  <mxGeometry x="20" y="60" width="160" height="40" as="geometry"/>
+</mxCell>
+<mxCell id="lane2" value="Backend" style="swimlane;" vertex="1" parent="1">
+  <mxGeometry x="280" y="40" width="200" height="200" as="geometry"/>
+</mxCell>
+<mxCell id="step2" value="Step 2" style="rounded=1;" vertex="1" parent="lane2">
+  <mxGeometry x="20" y="60" width="160" height="40" as="geometry"/>
+</mxCell>
+<mxCell id="edge1" style="edgeStyle=orthogonalEdgeStyle;endArrow=classic;" edge="1" parent="1" source="step1" target="step2">
+  <mxGeometry relative="1" as="geometry"/>
+</mxCell>
 
 Notes:
 - For AWS diagrams, use **AWS 2025 icons**.
@@ -335,38 +543,151 @@ Notes:
                 }),
             },
             edit_diagram: {
-                description: `Edit specific parts of the current diagram by replacing exact line matches. Use this tool to make targeted fixes without regenerating the entire XML.
-CRITICAL: Copy-paste the EXACT search pattern from the "Current diagram XML" in system context. Do NOT reorder attributes or reformat - the attribute order in draw.io XML varies and you MUST match it exactly.
-IMPORTANT: Keep edits concise:
-- COPY the exact mxCell line from the current XML (attribute order matters!)
-- Only include the lines that are changing, plus 1-2 surrounding lines for context if needed
-- Break large changes into multiple smaller edits
-- Each search must contain complete lines (never truncate mid-line)
-- First match only - be specific enough to target the right element`,
+                description: `Edit the current diagram by ID-based operations (update/add/delete cells).
+
+Operations:
+- update: Replace an existing cell by its id. Provide cell_id and complete new_xml.
+- add: Add a new cell. Provide cell_id (new unique id) and new_xml.
+- delete: Remove a cell by its id. Only cell_id is needed.
+
+For update/add, new_xml must be a complete mxCell element including mxGeometry.
+
+⚠️ JSON ESCAPING: Every " inside new_xml MUST be escaped as \\". Example: id=\\"5\\" value=\\"Label\\"`,
                 inputSchema: z.object({
-                    edits: z
+                    operations: z
                         .array(
                             z.object({
-                                search: z
+                                type: z
+                                    .enum(["update", "add", "delete"])
+                                    .describe("Operation type"),
+                                cell_id: z
                                     .string()
                                     .describe(
-                                        "EXACT lines copied from current XML (preserve attribute order!)",
+                                        "The id of the mxCell. Must match the id attribute in new_xml.",
                                     ),
-                                replace: z
+                                new_xml: z
                                     .string()
-                                    .describe("Replacement lines"),
+                                    .optional()
+                                    .describe(
+                                        "Complete mxCell XML element (required for update/add)",
+                                    ),
                             }),
                         )
+                        .describe("Array of operations to apply"),
+                }),
+            },
+            append_diagram: {
+                description: `Continue generating diagram XML when previous display_diagram output was truncated due to length limits.
+
+WHEN TO USE: Only call this tool after display_diagram was truncated (you'll see an error message about truncation).
+
+CRITICAL INSTRUCTIONS:
+1. Do NOT include any wrapper tags - just continue the mxCell elements
+2. Continue from EXACTLY where your previous output stopped
+3. Complete the remaining mxCell elements
+4. If still truncated, call append_diagram again with the next fragment
+
+Example: If previous output ended with '<mxCell id="x" style="rounded=1', continue with ';" vertex="1">...' and complete the remaining elements.`,
+                inputSchema: z.object({
+                    xml: z
+                        .string()
                         .describe(
-                            "Array of search/replace pairs to apply sequentially",
+                            "Continuation XML fragment to append (NO wrapper tags)",
                         ),
                 }),
             },
         },
-        temperature: 0,
+        ...(process.env.TEMPERATURE !== undefined && {
+            temperature: parseFloat(process.env.TEMPERATURE),
+        }),
     })
 
-    return result.toUIMessageStreamResponse()
+    return result.toUIMessageStreamResponse({
+        sendReasoning: true,
+        messageMetadata: ({ part }) => {
+            if (part.type === "finish") {
+                const usage = (part as any).totalUsage
+                if (!usage) {
+                    console.warn(
+                        "[messageMetadata] No usage data in finish part",
+                    )
+                    return undefined
+                }
+                // Total input = non-cached + cached (these are separate counts)
+                // Note: cacheWriteInputTokens is not available on finish part
+                const totalInputTokens =
+                    (usage.inputTokens ?? 0) + (usage.cachedInputTokens ?? 0)
+                return {
+                    inputTokens: totalInputTokens,
+                    outputTokens: usage.outputTokens ?? 0,
+                    finishReason: (part as any).finishReason,
+                }
+            }
+            return undefined
+        },
+    })
+}
+
+// Helper to categorize errors and return appropriate response
+function handleError(error: unknown): Response {
+    console.error("Error in chat route:", error)
+
+    const isDev = process.env.NODE_ENV === "development"
+
+    // Check for specific AI SDK error types
+    if (APICallError.isInstance(error)) {
+        return Response.json(
+            {
+                error: error.message,
+                ...(isDev && {
+                    details: error.responseBody,
+                    stack: error.stack,
+                }),
+            },
+            { status: error.statusCode || 500 },
+        )
+    }
+
+    if (LoadAPIKeyError.isInstance(error)) {
+        return Response.json(
+            {
+                error: "Authentication failed. Please check your API key.",
+                ...(isDev && {
+                    stack: error.stack,
+                }),
+            },
+            { status: 401 },
+        )
+    }
+
+    // Fallback for other errors with safety filter
+    const message =
+        error instanceof Error ? error.message : "An unexpected error occurred"
+    const status = (error as any)?.statusCode || (error as any)?.status || 500
+
+    // Prevent leaking API keys, tokens, or other sensitive data
+    const lowerMessage = message.toLowerCase()
+    const safeMessage =
+        lowerMessage.includes("key") ||
+        lowerMessage.includes("token") ||
+        lowerMessage.includes("sig") ||
+        lowerMessage.includes("signature") ||
+        lowerMessage.includes("secret") ||
+        lowerMessage.includes("password") ||
+        lowerMessage.includes("credential")
+            ? "Authentication failed. Please check your credentials."
+            : message
+
+    return Response.json(
+        {
+            error: safeMessage,
+            ...(isDev && {
+                details: message,
+                stack: error instanceof Error ? error.stack : undefined,
+            }),
+        },
+        { status },
+    )
 }
 
 // Wrap handler with error handling
@@ -374,11 +695,7 @@ async function safeHandler(req: Request): Promise<Response> {
     try {
         return await handleChatRequest(req)
     } catch (error) {
-        console.error("Error in chat route:", error)
-        return Response.json(
-            { error: "Internal server error" },
-            { status: 500 },
-        )
+        return handleError(error)
     }
 }
 

app/api/chat/xml_guide.md

@@ -81,16 +81,15 @@ Contains the actual diagram data.
 
 ## Root Cell Container: `<root>`
 
-Contains all the cells in the diagram.
+Contains all the cells in the diagram. **Note:** When generating diagrams, you only need to provide the mxCell elements - the root container and root cells (id="0", id="1") are added automatically.
 
-**Example:**
+**Internal structure (auto-generated):**
 
 ```xml
 <root>
-<mxCell id="0"/>
-<mxCell id="1" parent="0"/>
-
-  <!-- Other cells go here -->
+  <mxCell id="0"/>           <!-- Auto-added -->
+  <mxCell id="1" parent="0"/> <!-- Auto-added -->
+  <!-- Your mxCell elements go here (start from id="2") -->
 </root>
 ```
 
@@ -203,15 +202,15 @@ Draw.io files contain two special cells that are always present:
 1.  **Root Cell** (id = "0"): The parent of all cells
 2.  **Default Parent Cell** (id = "1", parent = "0"): The default layer and parent for most cells
 
-## Tips for Manually Creating Draw.io XML
+## Tips for Creating Draw.io XML
 
-1.  Start with the basic structure (`mxfile`, `diagram`, `mxGraphModel`, `root`)
-2.  Always include the two special cells (id = "0" and id = "1")
+1.  **Generate ONLY mxCell elements** - wrapper tags and root cells (id="0", id="1") are added automatically
+2.  Start IDs from "2" (id="0" and id="1" are reserved for root cells)
 3.  Assign unique and sequential IDs to all cells
-4.  Define parent relationships correctly
+4.  Define parent relationships correctly (use parent="1" for top-level shapes)
 5.  Use `mxGeometry` elements to position shapes
 6.  For connectors, specify `source` and `target` attributes
-7.  **CRITICAL: All mxCell elements must be DIRECT children of `<root>`. NEVER nest mxCell inside another mxCell.**
+7.  **CRITICAL: All mxCell elements must be siblings. NEVER nest mxCell inside another mxCell.**
 
 ## Common Patterns
 

app/api/config/route.ts

@@ -1,12 +1,10 @@
 import { NextResponse } from "next/server"
 
 export async function GET() {
-    const accessCodes =
-        process.env.ACCESS_CODE_LIST?.split(",")
-            .map((code) => code.trim())
-            .filter(Boolean) || []
-
     return NextResponse.json({
-        accessCodeRequired: accessCodes.length > 0,
+        accessCodeRequired: !!process.env.ACCESS_CODE_LIST,
+        dailyRequestLimit: Number(process.env.DAILY_REQUEST_LIMIT) || 0,
+        dailyTokenLimit: Number(process.env.DAILY_TOKEN_LIMIT) || 0,
+        tpmLimit: Number(process.env.TPM_LIMIT) || 0,
     })
 }

app/api/verify-access-code/route.ts

@@ -0,0 +1,32 @@
+export async function POST(req: Request) {
+    const accessCodes =
+        process.env.ACCESS_CODE_LIST?.split(",")
+            .map((code) => code.trim())
+            .filter(Boolean) || []
+
+    // If no access codes configured, verification always passes
+    if (accessCodes.length === 0) {
+        return Response.json({
+            valid: true,
+            message: "No access code required",
+        })
+    }
+
+    const accessCodeHeader = req.headers.get("x-access-code")
+
+    if (!accessCodeHeader) {
+        return Response.json(
+            { valid: false, message: "Access code is required" },
+            { status: 401 },
+        )
+    }
+
+    if (!accessCodes.includes(accessCodeHeader)) {
+        return Response.json(
+            { valid: false, message: "Invalid access code" },
+            { status: 401 },
+        )
+    }
+
+    return Response.json({ valid: true, message: "Access code is valid" })
+}

app/layout.tsx

@@ -1,5 +1,4 @@
 import { GoogleAnalytics } from "@next/third-parties/google"
-import { Analytics } from "@vercel/analytics/react"
 import type { Metadata, Viewport } from "next"
 import { JetBrains_Mono, Plus_Jakarta_Sans } from "next/font/google"
 import { DiagramProvider } from "@/contexts/diagram-context"
@@ -106,7 +105,7 @@ export default function RootLayout({
     }
 
     return (
-        <html lang="en">
+        <html lang="en" suppressHydrationWarning>
             <head>
                 <script
                     type="application/ld+json"
@@ -117,7 +116,6 @@ export default function RootLayout({
                 className={`${plusJakarta.variable} ${jetbrainsMono.variable} antialiased`}
             >
                 <DiagramProvider>{children}</DiagramProvider>
-                <Analytics />
             </body>
             {process.env.NEXT_PUBLIC_GA_ID && (
                 <GoogleAnalytics gaId={process.env.NEXT_PUBLIC_GA_ID} />

app/manifest.ts

@@ -0,0 +1,27 @@
+import type { MetadataRoute } from "next";
+
+export default function manifest(): MetadataRoute.Manifest {
+  return {
+    name: 'Next AI Draw.io',
+    short_name: 'AIDraw.io',
+    description: 'Create AWS architecture diagrams, flowcharts, and technical diagrams using AI. Free online tool integrating draw.io with AI assistance for professional diagram creation.',
+    start_url: '/',
+    display: 'standalone',
+    background_color: '#f9fafb',
+    theme_color: '#171d26',
+    icons: [
+      {
+        src: '/favicon-192x192.png',
+        sizes: '192x192',
+        type: 'image/png',
+        purpose: 'any',
+      },
+      {
+        src: '/favicon-512x512.png',
+        sizes: '512x512',
+        type: 'image/png',
+        purpose: 'any',
+      },
+    ],
+  }
+}

app/page.tsx

@@ -1,8 +1,9 @@
 "use client"
-import React, { useEffect, useRef, useState } from "react"
+import { useEffect, useRef, useState } from "react"
 import { DrawIoEmbed } from "react-drawio"
 import type { ImperativePanelHandle } from "react-resizable-panels"
 import ChatPanel from "@/components/chat-panel"
+import { STORAGE_CLOSE_PROTECTION_KEY } from "@/components/settings-dialog"
 import {
     ResizableHandle,
     ResizablePanel,
@@ -10,19 +11,74 @@ import {
 } from "@/components/ui/resizable"
 import { useDiagram } from "@/contexts/diagram-context"
 
+const drawioBaseUrl =
+    process.env.NEXT_PUBLIC_DRAWIO_BASE_URL || "https://embed.diagrams.net"
+
 export default function Home() {
-    const { drawioRef, handleDiagramExport } = useDiagram()
+    const {
+        drawioRef,
+        handleDiagramExport,
+        onDrawioLoad,
+        resetDrawioReady,
+        saveDiagramToStorage,
+    } = useDiagram()
     const [isMobile, setIsMobile] = useState(false)
     const [isChatVisible, setIsChatVisible] = useState(true)
-    const [drawioUi, setDrawioUi] = useState<"min" | "sketch">(() => {
-        if (typeof window !== "undefined") {
-            const saved = localStorage.getItem("drawio-theme")
-            if (saved === "min" || saved === "sketch") return saved
-        }
-        return "min"
-    })
+    const [drawioUi, setDrawioUi] = useState<"min" | "sketch">("min")
+    const [darkMode, setDarkMode] = useState(false)
+    const [isLoaded, setIsLoaded] = useState(false)
+    const [closeProtection, setCloseProtection] = useState(false)
+
     const chatPanelRef = useRef<ImperativePanelHandle>(null)
 
+    // Load preferences from localStorage after mount
+    useEffect(() => {
+        const savedUi = localStorage.getItem("drawio-theme")
+        if (savedUi === "min" || savedUi === "sketch") {
+            setDrawioUi(savedUi)
+        }
+
+        const savedDarkMode = localStorage.getItem("next-ai-draw-io-dark-mode")
+        if (savedDarkMode !== null) {
+            const isDark = savedDarkMode === "true"
+            setDarkMode(isDark)
+            document.documentElement.classList.toggle("dark", isDark)
+        } else {
+            const prefersDark = window.matchMedia(
+                "(prefers-color-scheme: dark)",
+            ).matches
+            setDarkMode(prefersDark)
+            document.documentElement.classList.toggle("dark", prefersDark)
+        }
+
+        const savedCloseProtection = localStorage.getItem(
+            STORAGE_CLOSE_PROTECTION_KEY,
+        )
+        if (savedCloseProtection === "true") {
+            setCloseProtection(true)
+        }
+
+        setIsLoaded(true)
+    }, [])
+
+    const handleDarkModeChange = async () => {
+        await saveDiagramToStorage()
+        const newValue = !darkMode
+        setDarkMode(newValue)
+        localStorage.setItem("next-ai-draw-io-dark-mode", String(newValue))
+        document.documentElement.classList.toggle("dark", newValue)
+        resetDrawioReady()
+    }
+
+    const handleDrawioUiChange = async () => {
+        await saveDiagramToStorage()
+        const newUi = drawioUi === "min" ? "sketch" : "min"
+        localStorage.setItem("drawio-theme", newUi)
+        setDrawioUi(newUi)
+        resetDrawioReady()
+    }
+
+    // Check mobile
     useEffect(() => {
         const checkMobile = () => {
             setIsMobile(window.innerWidth < 768)
@@ -46,6 +102,7 @@ export default function Home() {
         }
     }
 
+    // Keyboard shortcut for toggling chat panel
     useEffect(() => {
         const handleKeyDown = (event: KeyboardEvent) => {
             if ((event.ctrlKey || event.metaKey) && event.key === "b") {
@@ -59,8 +116,9 @@ export default function Home() {
     }, [])
 
     // Show confirmation dialog when user tries to leave the page
-    // This helps prevent accidental navigation from browser back gestures
     useEffect(() => {
+        if (!closeProtection) return
+
         const handleBeforeUnload = (event: BeforeUnloadEvent) => {
             event.preventDefault()
             return ""
@@ -69,35 +127,49 @@ export default function Home() {
         window.addEventListener("beforeunload", handleBeforeUnload)
         return () =>
             window.removeEventListener("beforeunload", handleBeforeUnload)
-    }, [])
+    }, [closeProtection])
 
     return (
         <div className="h-screen bg-background relative overflow-hidden">
             <ResizablePanelGroup
+                id="main-panel-group"
                 key={isMobile ? "mobile" : "desktop"}
                 direction={isMobile ? "vertical" : "horizontal"}
                 className="h-full"
             >
                 {/* Draw.io Canvas */}
-                <ResizablePanel defaultSize={isMobile ? 50 : 67} minSize={20}>
+                <ResizablePanel
+                    id="drawio-panel"
+                    defaultSize={isMobile ? 50 : 67}
+                    minSize={20}
+                >
                     <div
                         className={`h-full relative ${
                             isMobile ? "p-1" : "p-2"
                         }`}
                     >
-                        <div className="h-full rounded-xl overflow-hidden shadow-soft-lg border border-border/30 bg-white">
-                            <DrawIoEmbed
-                                key={drawioUi}
-                                ref={drawioRef}
-                                onExport={handleDiagramExport}
-                                urlParameters={{
-                                    ui: drawioUi,
-                                    spin: true,
-                                    libraries: false,
-                                    saveAndExit: false,
-                                    noExitBtn: true,
-                                }}
-                            />
+                        <div className="h-full rounded-xl overflow-hidden shadow-soft-lg border border-border/30">
+                            {isLoaded ? (
+                                <DrawIoEmbed
+                                    key={`${drawioUi}-${darkMode}`}
+                                    ref={drawioRef}
+                                    onExport={handleDiagramExport}
+                                    onLoad={onDrawioLoad}
+                                    baseUrl={drawioBaseUrl}
+                                    urlParameters={{
+                                        ui: drawioUi,
+                                        spin: true,
+                                        libraries: false,
+                                        saveAndExit: false,
+                                        noExitBtn: true,
+                                        dark: darkMode,
+                                    }}
+                                />
+                            ) : (
+                                <div className="h-full w-full flex items-center justify-center bg-background">
+                                    <div className="animate-spin h-8 w-8 border-4 border-primary border-t-transparent rounded-full" />
+                                </div>
+                            )}
                         </div>
                     </div>
                 </ResizablePanel>
@@ -106,6 +178,7 @@ export default function Home() {
 
                 {/* Chat Panel */}
                 <ResizablePanel
+                    id="chat-panel"
                     ref={chatPanelRef}
                     defaultSize={isMobile ? 50 : 33}
                     minSize={isMobile ? 20 : 15}
@@ -120,13 +193,11 @@ export default function Home() {
                             isVisible={isChatVisible}
                             onToggleVisibility={toggleChatPanel}
                             drawioUi={drawioUi}
-                            onToggleDrawioUi={() => {
-                                const newTheme =
-                                    drawioUi === "min" ? "sketch" : "min"
-                                localStorage.setItem("drawio-theme", newTheme)
-                                setDrawioUi(newTheme)
-                            }}
+                            onToggleDrawioUi={handleDrawioUiChange}
+                            darkMode={darkMode}
+                            onToggleDarkMode={handleDarkModeChange}
                             isMobile={isMobile}
+                            onCloseProtectionChange={setCloseProtection}
                         />
                     </div>
                 </ResizablePanel>

biome.json

@@ -19,6 +19,30 @@
             "recommended": true,
             "complexity": {
                 "noImportantStyles": "off"
+            },
+            "suspicious": {
+                "noExplicitAny": "off",
+                "noArrayIndexKey": "off",
+                "noImplicitAnyLet": "off",
+                "noAssignInExpressions": "off"
+            },
+            "a11y": {
+                "useButtonType": "off",
+                "noAutofocus": "off",
+                "noStaticElementInteractions": "off",
+                "useKeyWithClickEvents": "off",
+                "noLabelWithoutControl": "off",
+                "noNoninteractiveTabindex": "off"
+            },
+            "correctness": {
+                "useExhaustiveDependencies": "off"
+            },
+            "style": {
+                "useNodejsImportProtocol": "off",
+                "useTemplate": "off"
+            },
+            "security": {
+                "noDangerouslySetInnerHtml": "off"
             }
         }
     },

components/ai-elements/reasoning.tsx

@@ -0,0 +1,186 @@
+"use client"
+
+import { useControllableState } from "@radix-ui/react-use-controllable-state"
+import { BrainIcon, ChevronDownIcon } from "lucide-react"
+import type { ComponentProps, ReactNode } from "react"
+import { createContext, memo, useContext, useEffect, useState } from "react"
+import {
+    Collapsible,
+    CollapsibleContent,
+    CollapsibleTrigger,
+} from "@/components/ui/collapsible"
+import { cn } from "@/lib/utils"
+import { Shimmer } from "./shimmer"
+
+type ReasoningContextValue = {
+    isStreaming: boolean
+    isOpen: boolean
+    setIsOpen: (open: boolean) => void
+    duration: number | undefined
+}
+
+const ReasoningContext = createContext<ReasoningContextValue | null>(null)
+
+export const useReasoning = () => {
+    const context = useContext(ReasoningContext)
+    if (!context) {
+        throw new Error("Reasoning components must be used within Reasoning")
+    }
+    return context
+}
+
+export type ReasoningProps = ComponentProps<typeof Collapsible> & {
+    isStreaming?: boolean
+    open?: boolean
+    defaultOpen?: boolean
+    onOpenChange?: (open: boolean) => void
+    duration?: number
+}
+
+const AUTO_CLOSE_DELAY = 1000
+const MS_IN_S = 1000
+
+export const Reasoning = memo(
+    ({
+        className,
+        isStreaming = false,
+        open,
+        defaultOpen = true,
+        onOpenChange,
+        duration: durationProp,
+        children,
+        ...props
+    }: ReasoningProps) => {
+        const [isOpen, setIsOpen] = useControllableState({
+            prop: open,
+            defaultProp: defaultOpen,
+            onChange: onOpenChange,
+        })
+        const [duration, setDuration] = useControllableState({
+            prop: durationProp,
+            defaultProp: undefined,
+        })
+
+        const [hasAutoClosed, setHasAutoClosed] = useState(false)
+        const [startTime, setStartTime] = useState<number | null>(null)
+
+        // Track duration when streaming starts and ends
+        useEffect(() => {
+            if (isStreaming) {
+                if (startTime === null) {
+                    setStartTime(Date.now())
+                }
+            } else if (startTime !== null) {
+                setDuration(Math.ceil((Date.now() - startTime) / MS_IN_S))
+                setStartTime(null)
+            }
+        }, [isStreaming, startTime, setDuration])
+
+        // Auto-open when streaming starts, auto-close when streaming ends (once only)
+        useEffect(() => {
+            if (defaultOpen && !isStreaming && isOpen && !hasAutoClosed) {
+                // Add a small delay before closing to allow user to see the content
+                const timer = setTimeout(() => {
+                    setIsOpen(false)
+                    setHasAutoClosed(true)
+                }, AUTO_CLOSE_DELAY)
+
+                return () => clearTimeout(timer)
+            }
+        }, [isStreaming, isOpen, defaultOpen, setIsOpen, hasAutoClosed])
+
+        const handleOpenChange = (newOpen: boolean) => {
+            setIsOpen(newOpen)
+        }
+
+        return (
+            <ReasoningContext.Provider
+                value={{ isStreaming, isOpen, setIsOpen, duration }}
+            >
+                <Collapsible
+                    className={cn("not-prose mb-4", className)}
+                    onOpenChange={handleOpenChange}
+                    open={isOpen}
+                    {...props}
+                >
+                    {children}
+                </Collapsible>
+            </ReasoningContext.Provider>
+        )
+    },
+)
+
+export type ReasoningTriggerProps = ComponentProps<
+    typeof CollapsibleTrigger
+> & {
+    getThinkingMessage?: (isStreaming: boolean, duration?: number) => ReactNode
+}
+
+const defaultGetThinkingMessage = (isStreaming: boolean, duration?: number) => {
+    if (isStreaming || duration === 0) {
+        return <Shimmer duration={1}>Thinking...</Shimmer>
+    }
+    if (duration === undefined) {
+        return <p>Thought for a few seconds</p>
+    }
+    return <p>Thought for {duration} seconds</p>
+}
+
+export const ReasoningTrigger = memo(
+    ({
+        className,
+        children,
+        getThinkingMessage = defaultGetThinkingMessage,
+        ...props
+    }: ReasoningTriggerProps) => {
+        const { isStreaming, isOpen, duration } = useReasoning()
+
+        return (
+            <CollapsibleTrigger
+                className={cn(
+                    "flex w-full items-center gap-2 text-muted-foreground text-sm transition-colors hover:text-foreground",
+                    className,
+                )}
+                {...props}
+            >
+                {children ?? (
+                    <>
+                        <BrainIcon className="size-4" />
+                        {getThinkingMessage(isStreaming, duration)}
+                        <ChevronDownIcon
+                            className={cn(
+                                "size-4 transition-transform",
+                                isOpen ? "rotate-180" : "rotate-0",
+                            )}
+                        />
+                    </>
+                )}
+            </CollapsibleTrigger>
+        )
+    },
+)
+
+export type ReasoningContentProps = ComponentProps<
+    typeof CollapsibleContent
+> & {
+    children: string
+}
+
+export const ReasoningContent = memo(
+    ({ className, children, ...props }: ReasoningContentProps) => (
+        <CollapsibleContent
+            className={cn(
+                "mt-4 text-sm",
+                "data-[state=closed]:fade-out-0 data-[state=closed]:slide-out-to-top-2 data-[state=open]:slide-in-from-top-2 text-muted-foreground outline-none data-[state=closed]:animate-out data-[state=open]:animate-in",
+                className,
+            )}
+            {...props}
+        >
+            <div className="whitespace-pre-wrap">{children}</div>
+        </CollapsibleContent>
+    ),
+)
+
+Reasoning.displayName = "Reasoning"
+ReasoningTrigger.displayName = "ReasoningTrigger"
+ReasoningContent.displayName = "ReasoningContent"

components/ai-elements/shimmer.tsx

@@ -0,0 +1,64 @@
+"use client"
+
+import { motion } from "motion/react"
+import {
+    type CSSProperties,
+    type ElementType,
+    type JSX,
+    memo,
+    useMemo,
+} from "react"
+import { cn } from "@/lib/utils"
+
+export type TextShimmerProps = {
+    children: string
+    as?: ElementType
+    className?: string
+    duration?: number
+    spread?: number
+}
+
+const ShimmerComponent = ({
+    children,
+    as: Component = "p",
+    className,
+    duration = 2,
+    spread = 2,
+}: TextShimmerProps) => {
+    const MotionComponent = motion.create(
+        Component as keyof JSX.IntrinsicElements,
+    )
+
+    const dynamicSpread = useMemo(
+        () => (children?.length ?? 0) * spread,
+        [children, spread],
+    )
+
+    return (
+        <MotionComponent
+            animate={{ backgroundPosition: "0% center" }}
+            className={cn(
+                "relative inline-block bg-[length:250%_100%,auto] bg-clip-text text-transparent",
+                "[--bg:linear-gradient(90deg,#0000_calc(50%-var(--spread)),var(--color-background),#0000_calc(50%+var(--spread)))] [background-repeat:no-repeat,padding-box]",
+                className,
+            )}
+            initial={{ backgroundPosition: "100% center" }}
+            style={
+                {
+                    "--spread": `${dynamicSpread}px`,
+                    backgroundImage:
+                        "var(--bg), linear-gradient(var(--color-muted-foreground), var(--color-muted-foreground))",
+                } as CSSProperties
+            }
+            transition={{
+                repeat: Number.POSITIVE_INFINITY,
+                duration,
+                ease: "linear",
+            }}
+        >
+            {children}
+        </MotionComponent>
+    )
+}
+
+export const Shimmer = memo(ShimmerComponent)

components/chat-example-panel.tsx

@@ -1,28 +1,59 @@
 "use client"
 
-import { Cloud, GitBranch, Palette, Zap } from "lucide-react"
+import {
+    Cloud,
+    FileText,
+    GitBranch,
+    Palette,
+    Terminal,
+    Zap,
+} from "lucide-react"
 
 interface ExampleCardProps {
     icon: React.ReactNode
     title: string
     description: string
     onClick: () => void
+    isNew?: boolean
 }
 
-function ExampleCard({ icon, title, description, onClick }: ExampleCardProps) {
+function ExampleCard({
+    icon,
+    title,
+    description,
+    onClick,
+    isNew,
+}: ExampleCardProps) {
     return (
         <button
             onClick={onClick}
-            className="group w-full text-left p-4 rounded-xl border border-border/60 bg-card hover:bg-accent/50 hover:border-primary/30 transition-all duration-200 hover:shadow-sm"
+            className={`group w-full text-left p-4 rounded-xl border bg-card hover:bg-accent/50 hover:border-primary/30 transition-all duration-200 hover:shadow-sm ${
+                isNew
+                    ? "border-primary/40 ring-1 ring-primary/20"
+                    : "border-border/60"
+            }`}
         >
             <div className="flex items-start gap-3">
-                <div className="w-9 h-9 rounded-lg bg-primary/10 flex items-center justify-center shrink-0 group-hover:bg-primary/15 transition-colors">
+                <div
+                    className={`w-9 h-9 rounded-lg flex items-center justify-center shrink-0 transition-colors ${
+                        isNew
+                            ? "bg-primary/20 group-hover:bg-primary/25"
+                            : "bg-primary/10 group-hover:bg-primary/15"
+                    }`}
+                >
                     {icon}
                 </div>
                 <div className="min-w-0">
-                    <h3 className="text-sm font-medium text-foreground group-hover:text-primary transition-colors">
-                        {title}
-                    </h3>
+                    <div className="flex items-center gap-2">
+                        <h3 className="text-sm font-medium text-foreground group-hover:text-primary transition-colors">
+                            {title}
+                        </h3>
+                        {isNew && (
+                            <span className="px-1.5 py-0.5 text-[10px] font-semibold bg-primary text-primary-foreground rounded">
+                                NEW
+                            </span>
+                        )}
+                    </div>
                     <p className="text-xs text-muted-foreground mt-0.5 line-clamp-2">
                         {description}
                     </p>
@@ -67,8 +98,50 @@ export default function ExamplePanel({
         }
     }
 
+    const handlePdfExample = async () => {
+        setInput("Summarize this paper as a diagram")
+
+        try {
+            const response = await fetch("/chain-of-thought.txt")
+            const blob = await response.blob()
+            const file = new File([blob], "chain-of-thought.txt", {
+                type: "text/plain",
+            })
+            setFiles([file])
+        } catch (error) {
+            console.error("Error loading text file:", error)
+        }
+    }
+
     return (
         <div className="py-6 px-2 animate-fade-in">
+            {/* MCP Server Notice */}
+            <a
+                href="https://github.com/DayuanJiang/next-ai-draw-io/tree/main/packages/mcp-server"
+                target="_blank"
+                rel="noopener noreferrer"
+                className="block mb-4 p-3 rounded-xl bg-gradient-to-r from-purple-500/10 to-blue-500/10 border border-purple-500/20 hover:border-purple-500/40 transition-colors group"
+            >
+                <div className="flex items-center gap-3">
+                    <div className="w-8 h-8 rounded-lg bg-purple-500/20 flex items-center justify-center shrink-0">
+                        <Terminal className="w-4 h-4 text-purple-500" />
+                    </div>
+                    <div className="min-w-0">
+                        <div className="flex items-center gap-2">
+                            <span className="text-sm font-medium text-foreground group-hover:text-purple-500 transition-colors">
+                                MCP Server
+                            </span>
+                            <span className="px-1.5 py-0.5 text-[10px] font-semibold bg-purple-500 text-white rounded">
+                                PREVIEW
+                            </span>
+                        </div>
+                        <p className="text-xs text-muted-foreground">
+                            Use in Claude Desktop, VS Code & Cursor
+                        </p>
+                    </div>
+                </div>
+            </a>
+
             {/* Welcome section */}
             <div className="text-center mb-6">
                 <h2 className="text-lg font-semibold text-foreground mb-2">
@@ -87,6 +160,14 @@ export default function ExamplePanel({
                 </p>
 
                 <div className="grid gap-2">
+                    <ExampleCard
+                        icon={<FileText className="w-4 h-4 text-primary" />}
+                        title="Paper to Diagram"
+                        description="Upload .pdf, .txt, .md, .json, .csv, .py, .js, .ts and more"
+                        onClick={handlePdfExample}
+                        isNew
+                    />
+
                     <ExampleCard
                         icon={<Zap className="w-4 h-4 text-primary" />}
                         title="Animated Diagram"

components/chat-input.tsx

@@ -4,9 +4,7 @@ import {
     Download,
     History,
     Image as ImageIcon,
-    LayoutGrid,
     Loader2,
-    PenTool,
     Send,
     Trash2,
 } from "lucide-react"
@@ -19,21 +17,24 @@ import { HistoryDialog } from "@/components/history-dialog"
 import { ResetWarningModal } from "@/components/reset-warning-modal"
 import { SaveDialog } from "@/components/save-dialog"
 import { Button } from "@/components/ui/button"
-import {
-    Dialog,
-    DialogContent,
-    DialogDescription,
-    DialogFooter,
-    DialogHeader,
-    DialogTitle,
-} from "@/components/ui/dialog"
+import { Switch } from "@/components/ui/switch"
 import { Textarea } from "@/components/ui/textarea"
+import {
+    Tooltip,
+    TooltipContent,
+    TooltipTrigger,
+} from "@/components/ui/tooltip"
 import { useDiagram } from "@/contexts/diagram-context"
+import { isPdfFile, isTextFile } from "@/lib/pdf-utils"
 import { FilePreviewList } from "./file-preview-list"
 
-const MAX_FILE_SIZE = 2 * 1024 * 1024 // 2MB
+const MAX_IMAGE_SIZE = 2 * 1024 * 1024 // 2MB
 const MAX_FILES = 5
 
+function isValidFileType(file: File): boolean {
+    return file.type.startsWith("image/") || isPdfFile(file) || isTextFile(file)
+}
+
 function formatFileSize(bytes: number): string {
     const mb = bytes / 1024 / 1024
     if (mb < 0.01) return `${(bytes / 1024).toFixed(0)}KB`
@@ -73,9 +74,16 @@ function validateFiles(
             errors.push(`Only ${availableSlots} more file(s) allowed`)
             break
         }
-        if (file.size > MAX_FILE_SIZE) {
+        if (!isValidFileType(file)) {
+            errors.push(`"${file.name}" is not a supported file type`)
+            continue
+        }
+        // Only check size for images (PDFs/text files are extracted client-side, so file size doesn't matter)
+        const isExtractedFile = isPdfFile(file) || isTextFile(file)
+        if (!isExtractedFile && file.size > MAX_IMAGE_SIZE) {
+            const maxSizeMB = MAX_IMAGE_SIZE / 1024 / 1024
             errors.push(
-                `"${file.name}" is ${formatFileSize(file.size)} (exceeds 2MB)`,
+                `"${file.name}" is ${formatFileSize(file.size)} (exceeds ${maxSizeMB}MB)`,
             )
         } else {
             validFiles.push(file)
@@ -99,8 +107,8 @@ function showValidationErrors(errors: string[]) {
                     {errors.length} files rejected:
                 </span>
                 <ul className="text-muted-foreground text-xs list-disc list-inside">
-                    {errors.slice(0, 3).map((err, i) => (
-                        <li key={i}>{err}</li>
+                    {errors.slice(0, 3).map((err) => (
+                        <li key={err}>{err}</li>
                     ))}
                     {errors.length > 3 && (
                         <li>...and {errors.length - 3} more</li>
@@ -119,12 +127,16 @@ interface ChatInputProps {
     onClearChat: () => void
     files?: File[]
     onFileChange?: (files: File[]) => void
+    pdfData?: Map<
+        File,
+        { text: string; charCount: number; isExtracting: boolean }
+    >
     showHistory?: boolean
     onToggleHistory?: (show: boolean) => void
     sessionId?: string
     error?: Error | null
-    drawioUi?: "min" | "sketch"
-    onToggleDrawioUi?: () => void
+    minimalStyle?: boolean
+    onMinimalStyleChange?: (value: boolean) => void
 }
 
 export function ChatInput({
@@ -135,12 +147,13 @@ export function ChatInput({
     onClearChat,
     files = [],
     onFileChange = () => {},
+    pdfData = new Map(),
     showHistory = false,
     onToggleHistory = () => {},
     sessionId,
     error = null,
-    drawioUi = "min",
-    onToggleDrawioUi = () => {},
+    minimalStyle = false,
+    onMinimalStyleChange = () => {},
 }: ChatInputProps) {
     const { diagramHistory, saveDiagramToFile } = useDiagram()
     const textareaRef = useRef<HTMLTextAreaElement>(null)
@@ -148,7 +161,6 @@ export function ChatInput({
     const [isDragging, setIsDragging] = useState(false)
     const [showClearDialog, setShowClearDialog] = useState(false)
     const [showSaveDialog, setShowSaveDialog] = useState(false)
-    const [showThemeWarning, setShowThemeWarning] = useState(false)
 
     // Allow retry when there's an error (even if status is still "streaming" or "submitted")
     const isDisabled =
@@ -162,10 +174,16 @@ export function ChatInput({
         }
     }, [])
 
+    // Handle programmatic input changes (e.g., setInput("") after form submission)
     useEffect(() => {
         adjustTextareaHeight()
     }, [input, adjustTextareaHeight])
 
+    const handleChange = (e: React.ChangeEvent<HTMLTextAreaElement>) => {
+        onChange(e)
+        adjustTextareaHeight()
+    }
+
     const handleKeyDown = (e: React.KeyboardEvent) => {
         if ((e.metaKey || e.ctrlKey) && e.key === "Enter") {
             e.preventDefault()
@@ -254,11 +272,14 @@ export function ChatInput({
         if (isDisabled) return
 
         const droppedFiles = e.dataTransfer.files
-        const imageFiles = Array.from(droppedFiles).filter((file) =>
-            file.type.startsWith("image/"),
+        const supportedFiles = Array.from(droppedFiles).filter((file) =>
+            isValidFileType(file),
         )
 
-        const { validFiles, errors } = validateFiles(imageFiles, files.length)
+        const { validFiles, errors } = validateFiles(
+            supportedFiles,
+            files.length,
+        )
         showValidationErrors(errors)
         if (validFiles.length > 0) {
             onFileChange([...files, ...validFiles])
@@ -288,6 +309,7 @@ export function ChatInput({
                     <FilePreviewList
                         files={files}
                         onRemoveFile={handleRemoveFile}
+                        pdfData={pdfData}
                     />
                 </div>
             )}
@@ -297,10 +319,10 @@ export function ChatInput({
                 <Textarea
                     ref={textareaRef}
                     value={input}
-                    onChange={onChange}
+                    onChange={handleChange}
                     onKeyDown={handleKeyDown}
                     onPaste={handlePaste}
-                    placeholder="Describe your diagram or paste an image..."
+                    placeholder="Describe your diagram or upload a file..."
                     disabled={isDisabled}
                     aria-label="Chat input"
                     className="min-h-[60px] max-h-[200px] resize-none border-0 bg-transparent px-4 py-3 text-sm focus-visible:ring-0 focus-visible:ring-offset-0 placeholder:text-muted-foreground/60"
@@ -332,59 +354,31 @@ export function ChatInput({
                             onToggleHistory={onToggleHistory}
                         />
 
-                        <ButtonWithTooltip
-                            type="button"
-                            variant="ghost"
-                            size="sm"
-                            onClick={() => setShowThemeWarning(true)}
-                            tooltipContent={
-                                drawioUi === "min"
-                                    ? "Switch to Sketch theme"
-                                    : "Switch to Minimal theme"
-                            }
-                            className="h-8 w-8 p-0 text-muted-foreground hover:text-foreground"
-                        >
-                            {drawioUi === "min" ? (
-                                <PenTool className="h-4 w-4" />
-                            ) : (
-                                <LayoutGrid className="h-4 w-4" />
-                            )}
-                        </ButtonWithTooltip>
-
-                        <Dialog
-                            open={showThemeWarning}
-                            onOpenChange={setShowThemeWarning}
-                        >
-                            <DialogContent>
-                                <DialogHeader>
-                                    <DialogTitle>Switch Theme?</DialogTitle>
-                                    <DialogDescription>
-                                        Switching themes will reload the diagram
-                                        editor and clear any unsaved changes.
-                                    </DialogDescription>
-                                </DialogHeader>
-                                <DialogFooter>
-                                    <Button
-                                        variant="outline"
-                                        onClick={() =>
-                                            setShowThemeWarning(false)
-                                        }
+                        <Tooltip>
+                            <TooltipTrigger asChild>
+                                <div className="flex items-center gap-1.5">
+                                    <Switch
+                                        id="minimal-style"
+                                        checked={minimalStyle}
+                                        onCheckedChange={onMinimalStyleChange}
+                                        className="scale-75"
+                                    />
+                                    <label
+                                        htmlFor="minimal-style"
+                                        className={`text-xs cursor-pointer select-none ${
+                                            minimalStyle
+                                                ? "text-primary font-medium"
+                                                : "text-muted-foreground"
+                                        }`}
                                     >
-                                        Cancel
-                                    </Button>
-                                    <Button
-                                        variant="destructive"
-                                        onClick={() => {
-                                            onClearChat()
-                                            onToggleDrawioUi()
-                                            setShowThemeWarning(false)
-                                        }}
-                                    >
-                                        Switch Theme
-                                    </Button>
-                                </DialogFooter>
-                            </DialogContent>
-                        </Dialog>
+                                        {minimalStyle ? "Minimal" : "Styled"}
+                                    </label>
+                                </div>
+                            </TooltipTrigger>
+                            <TooltipContent side="top">
+                                Use minimal for faster generation (no colors)
+                            </TooltipContent>
+                        </Tooltip>
                     </div>
 
                     {/* Right actions */}
@@ -430,7 +424,7 @@ export function ChatInput({
                             size="sm"
                             onClick={triggerFileInput}
                             disabled={isDisabled}
-                            tooltipContent="Upload image"
+                            tooltipContent="Upload file (image, PDF, text)"
                             className="h-8 w-8 p-0 text-muted-foreground hover:text-foreground"
                         >
                             <ImageIcon className="h-4 w-4" />
@@ -441,7 +435,7 @@ export function ChatInput({
                             ref={fileInputRef}
                             className="hidden"
                             onChange={handleFileChange}
-                            accept="image/*"
+                            accept="image/*,.pdf,application/pdf,text/*,.md,.markdown,.json,.csv,.xml,.yaml,.yml,.toml"
                             multiple
                             disabled={isDisabled}
                         />

components/code-block.tsx

@@ -12,7 +12,7 @@ export function CodeBlock({ code, language = "xml" }: CodeBlockProps) {
         <div className="overflow-hidden w-full">
             <Highlight theme={themes.github} code={code} language={language}>
                 {({
-                    className,
+                    className: _className,
                     style,
                     tokens,
                     getLineProps,

components/file-preview-list.tsx

@@ -1,53 +1,142 @@
 "use client"
 
-import { X } from "lucide-react"
+import { FileCode, FileText, Loader2, X } from "lucide-react"
 import Image from "next/image"
-import React, { useEffect, useState } from "react"
+import { useEffect, useRef, useState } from "react"
+import { isPdfFile, isTextFile } from "@/lib/pdf-utils"
+
+function formatCharCount(count: number): string {
+    if (count >= 1000) {
+        return `${(count / 1000).toFixed(1)}k`
+    }
+    return String(count)
+}
 
 interface FilePreviewListProps {
     files: File[]
     onRemoveFile: (fileToRemove: File) => void
+    pdfData?: Map<
+        File,
+        { text: string; charCount: number; isExtracting: boolean }
+    >
 }
 
-export function FilePreviewList({ files, onRemoveFile }: FilePreviewListProps) {
+export function FilePreviewList({
+    files,
+    onRemoveFile,
+    pdfData = new Map(),
+}: FilePreviewListProps) {
     const [selectedImage, setSelectedImage] = useState<string | null>(null)
+    const [imageUrls, setImageUrls] = useState<Map<File, string>>(new Map())
+    const imageUrlsRef = useRef<Map<File, string>>(new Map())
 
-    // Cleanup object URLs on unmount
+    // Create and cleanup object URLs when files change
     useEffect(() => {
-        const objectUrls = files
-            .filter((file) => file.type.startsWith("image/"))
-            .map((file) => URL.createObjectURL(file))
+        const currentUrls = imageUrlsRef.current
+        const newUrls = new Map<File, string>()
 
-        return () => {
-            objectUrls.forEach(URL.revokeObjectURL)
-        }
+        files.forEach((file) => {
+            if (file.type.startsWith("image/")) {
+                // Reuse existing URL if file is already tracked
+                const existingUrl = currentUrls.get(file)
+                if (existingUrl) {
+                    newUrls.set(file, existingUrl)
+                } else {
+                    newUrls.set(file, URL.createObjectURL(file))
+                }
+            }
+        })
+
+        // Revoke URLs for files that are no longer in the list
+        currentUrls.forEach((url, file) => {
+            if (!newUrls.has(file)) {
+                URL.revokeObjectURL(url)
+            }
+        })
+
+        imageUrlsRef.current = newUrls
+        setImageUrls(newUrls)
     }, [files])
 
+    // Cleanup all URLs on unmount only
+    useEffect(() => {
+        return () => {
+            imageUrlsRef.current.forEach((url) => {
+                URL.revokeObjectURL(url)
+            })
+            // Clear the ref so StrictMode remount creates fresh URLs
+            imageUrlsRef.current = new Map()
+        }
+    }, [])
+
+    // Clear selected image if its URL was revoked
+    useEffect(() => {
+        if (
+            selectedImage &&
+            !Array.from(imageUrls.values()).includes(selectedImage)
+        ) {
+            setSelectedImage(null)
+        }
+    }, [imageUrls, selectedImage])
+
     if (files.length === 0) return null
 
     return (
         <>
             <div className="flex flex-wrap gap-2 mt-2 p-2 bg-muted/50 rounded-md">
                 {files.map((file, index) => {
-                    const imageUrl = file.type.startsWith("image/")
-                        ? URL.createObjectURL(file)
-                        : null
+                    const imageUrl = imageUrls.get(file) || null
+                    const pdfInfo = pdfData.get(file)
                     return (
                         <div key={file.name + index} className="relative group">
                             <div
-                                className="w-20 h-20 border rounded-md overflow-hidden bg-muted cursor-pointer"
+                                className={`w-20 h-20 border rounded-md overflow-hidden bg-muted ${
+                                    file.type.startsWith("image/") && imageUrl
+                                        ? "cursor-pointer"
+                                        : ""
+                                }`}
                                 onClick={() =>
-                                    imageUrl && setSelectedImage(imageUrl)
+                                    file.type.startsWith("image/") &&
+                                    imageUrl &&
+                                    setSelectedImage(imageUrl)
                                 }
                             >
-                                {file.type.startsWith("image/") ? (
+                                {file.type.startsWith("image/") && imageUrl ? (
                                     <Image
-                                        src={imageUrl!}
+                                        src={imageUrl}
                                         alt={file.name}
                                         width={80}
                                         height={80}
                                         className="object-cover w-full h-full"
+                                        unoptimized
                                     />
+                                ) : isPdfFile(file) || isTextFile(file) ? (
+                                    <div className="flex flex-col items-center justify-center h-full p-1">
+                                        {pdfInfo?.isExtracting ? (
+                                            <Loader2 className="h-6 w-6 text-blue-500 mb-1 animate-spin" />
+                                        ) : isPdfFile(file) ? (
+                                            <FileText className="h-6 w-6 text-red-500 mb-1" />
+                                        ) : (
+                                            <FileCode className="h-6 w-6 text-blue-500 mb-1" />
+                                        )}
+                                        <span className="text-xs text-center truncate w-full px-1">
+                                            {file.name.length > 10
+                                                ? `${file.name.slice(0, 7)}...`
+                                                : file.name}
+                                        </span>
+                                        {pdfInfo?.isExtracting ? (
+                                            <span className="text-[10px] text-muted-foreground">
+                                                Reading...
+                                            </span>
+                                        ) : pdfInfo?.charCount ? (
+                                            <span className="text-[10px] text-green-600 font-medium">
+                                                {formatCharCount(
+                                                    pdfInfo.charCount,
+                                                )}{" "}
+                                                chars
+                                            </span>
+                                        ) : null}
+                                    </div>
                                 ) : (
                                     <div className="flex items-center justify-center h-full text-xs text-center p-1">
                                         {file.name}
@@ -88,6 +177,7 @@ export function FilePreviewList({ files, onRemoveFile }: FilePreviewListProps) {
                             height={900}
                             className="object-contain max-w-full max-h-[90vh] w-auto h-auto"
                             onClick={(e) => e.stopPropagation()}
+                            unoptimized
                         />
                     </div>
                 </div>

components/history-dialog.tsx

@@ -32,7 +32,8 @@ export function HistoryDialog({
 
     const handleConfirmRestore = () => {
         if (selectedIndex !== null) {
-            onDisplayChart(diagramHistory[selectedIndex].xml)
+            // Skip validation for trusted history snapshots
+            onDisplayChart(diagramHistory[selectedIndex].xml, true)
             handleClose()
         }
     }

components/quota-limit-toast.tsx

@@ -0,0 +1,115 @@
+"use client"
+
+import { Coffee, X } from "lucide-react"
+import Link from "next/link"
+import type React from "react"
+import { FaGithub } from "react-icons/fa"
+
+interface QuotaLimitToastProps {
+    type?: "request" | "token"
+    used: number
+    limit: number
+    onDismiss: () => void
+}
+
+export function QuotaLimitToast({
+    type = "request",
+    used,
+    limit,
+    onDismiss,
+}: QuotaLimitToastProps) {
+    const isTokenLimit = type === "token"
+    const formatNumber = (n: number) =>
+        n >= 1000 ? `${(n / 1000).toFixed(1)}k` : n.toString()
+    const handleKeyDown = (e: React.KeyboardEvent) => {
+        if (e.key === "Escape") {
+            e.preventDefault()
+            onDismiss()
+        }
+    }
+
+    return (
+        <div
+            role="alert"
+            aria-live="polite"
+            tabIndex={0}
+            onKeyDown={handleKeyDown}
+            className="relative w-[400px] overflow-hidden rounded-xl border border-border/50 bg-card p-5 shadow-soft animate-message-in"
+        >
+            {/* Close button */}
+            <button
+                onClick={onDismiss}
+                className="absolute right-3 top-3 p-1.5 rounded-full text-muted-foreground/60 hover:text-foreground hover:bg-muted transition-colors"
+                aria-label="Dismiss"
+            >
+                <X className="w-4 h-4" />
+            </button>
+
+            {/* Title row with icon */}
+            <div className="flex items-center gap-2.5 mb-3 pr-6">
+                <div className="flex-shrink-0 w-8 h-8 rounded-lg bg-accent flex items-center justify-center">
+                    <Coffee
+                        className="w-4 h-4 text-accent-foreground"
+                        strokeWidth={2}
+                    />
+                </div>
+                <h3 className="font-semibold text-foreground text-sm">
+                    {isTokenLimit
+                        ? "Daily Token Limit Reached"
+                        : "Daily Quota Reached"}
+                </h3>
+                <span className="px-2 py-0.5 text-xs font-medium rounded-md bg-muted text-muted-foreground">
+                    {isTokenLimit
+                        ? `${formatNumber(used)}/${formatNumber(limit)} tokens`
+                        : `${used}/${limit}`}
+                </span>
+            </div>
+
+            {/* Message */}
+            <div className="text-sm text-muted-foreground leading-relaxed mb-4 space-y-2">
+                <p>
+                    Oops — you've reached the daily{" "}
+                    {isTokenLimit ? "token" : "API"} limit for this demo! As an
+                    indie developer covering all the API costs myself, I have to
+                    set these limits to keep things sustainable.{" "}
+                    <Link
+                        href="/about"
+                        target="_blank"
+                        rel="noopener noreferrer"
+                        className="inline-flex items-center gap-1 text-amber-600 font-medium hover:text-amber-700 hover:underline"
+                    >
+                        Learn more →
+                    </Link>
+                </p>
+                <p>
+                    <strong>Tip:</strong> You can use your own API key (click
+                    the Settings icon) or self-host the project to bypass these
+                    limits.
+                </p>
+                <p>Your limit resets tomorrow. Thanks for understanding!</p>
+            </div>
+
+            {/* Action buttons */}
+            <div className="flex items-center gap-2">
+                <a
+                    href="https://github.com/DayuanJiang/next-ai-draw-io"
+                    target="_blank"
+                    rel="noopener noreferrer"
+                    className="inline-flex items-center gap-1.5 px-3 py-1.5 text-xs font-medium rounded-lg bg-primary text-primary-foreground hover:bg-primary/90 transition-colors"
+                >
+                    <FaGithub className="w-3.5 h-3.5" />
+                    Self-host
+                </a>
+                <a
+                    href="https://github.com/sponsors/DayuanJiang"
+                    target="_blank"
+                    rel="noopener noreferrer"
+                    className="inline-flex items-center gap-1.5 px-3 py-1.5 text-xs font-medium rounded-lg border border-border text-foreground hover:bg-muted transition-colors"
+                >
+                    <Coffee className="w-3.5 h-3.5" />
+                    Sponsor
+                </a>
+            </div>
+        </div>
+    )
+}

components/settings-dialog.tsx

@@ -1,38 +1,145 @@
 "use client"
 
+import { Moon, Sun } from "lucide-react"
 import { useEffect, useState } from "react"
 import { Button } from "@/components/ui/button"
 import {
     Dialog,
     DialogContent,
     DialogDescription,
-    DialogFooter,
     DialogHeader,
     DialogTitle,
 } from "@/components/ui/dialog"
 import { Input } from "@/components/ui/input"
+import { Label } from "@/components/ui/label"
+import {
+    Select,
+    SelectContent,
+    SelectItem,
+    SelectTrigger,
+    SelectValue,
+} from "@/components/ui/select"
+import { Switch } from "@/components/ui/switch"
 
 interface SettingsDialogProps {
     open: boolean
     onOpenChange: (open: boolean) => void
+    onCloseProtectionChange?: (enabled: boolean) => void
+    drawioUi: "min" | "sketch"
+    onToggleDrawioUi: () => void
+    darkMode: boolean
+    onToggleDarkMode: () => void
 }
 
 export const STORAGE_ACCESS_CODE_KEY = "next-ai-draw-io-access-code"
+export const STORAGE_CLOSE_PROTECTION_KEY = "next-ai-draw-io-close-protection"
+const STORAGE_ACCESS_CODE_REQUIRED_KEY = "next-ai-draw-io-access-code-required"
+export const STORAGE_AI_PROVIDER_KEY = "next-ai-draw-io-ai-provider"
+export const STORAGE_AI_BASE_URL_KEY = "next-ai-draw-io-ai-base-url"
+export const STORAGE_AI_API_KEY_KEY = "next-ai-draw-io-ai-api-key"
+export const STORAGE_AI_MODEL_KEY = "next-ai-draw-io-ai-model"
 
-export function SettingsDialog({ open, onOpenChange }: SettingsDialogProps) {
+function getStoredAccessCodeRequired(): boolean | null {
+    if (typeof window === "undefined") return null
+    const stored = localStorage.getItem(STORAGE_ACCESS_CODE_REQUIRED_KEY)
+    if (stored === null) return null
+    return stored === "true"
+}
+
+export function SettingsDialog({
+    open,
+    onOpenChange,
+    onCloseProtectionChange,
+    drawioUi,
+    onToggleDrawioUi,
+    darkMode,
+    onToggleDarkMode,
+}: SettingsDialogProps) {
     const [accessCode, setAccessCode] = useState("")
+    const [closeProtection, setCloseProtection] = useState(true)
+    const [isVerifying, setIsVerifying] = useState(false)
+    const [error, setError] = useState("")
+    const [accessCodeRequired, setAccessCodeRequired] = useState(
+        () => getStoredAccessCodeRequired() ?? false,
+    )
+    const [provider, setProvider] = useState("")
+    const [baseUrl, setBaseUrl] = useState("")
+    const [apiKey, setApiKey] = useState("")
+    const [modelId, setModelId] = useState("")
+
+    useEffect(() => {
+        // Only fetch if not cached in localStorage
+        if (getStoredAccessCodeRequired() !== null) return
+
+        fetch("/api/config")
+            .then((res) => {
+                if (!res.ok) throw new Error(`HTTP ${res.status}`)
+                return res.json()
+            })
+            .then((data) => {
+                const required = data?.accessCodeRequired === true
+                localStorage.setItem(
+                    STORAGE_ACCESS_CODE_REQUIRED_KEY,
+                    String(required),
+                )
+                setAccessCodeRequired(required)
+            })
+            .catch(() => {
+                // Don't cache on error - allow retry on next mount
+                setAccessCodeRequired(false)
+            })
+    }, [])
 
     useEffect(() => {
         if (open) {
             const storedCode =
                 localStorage.getItem(STORAGE_ACCESS_CODE_KEY) || ""
             setAccessCode(storedCode)
+
+            const storedCloseProtection = localStorage.getItem(
+                STORAGE_CLOSE_PROTECTION_KEY,
+            )
+            // Default to true if not set
+            setCloseProtection(storedCloseProtection !== "false")
+
+            // Load AI provider settings
+            setProvider(localStorage.getItem(STORAGE_AI_PROVIDER_KEY) || "")
+            setBaseUrl(localStorage.getItem(STORAGE_AI_BASE_URL_KEY) || "")
+            setApiKey(localStorage.getItem(STORAGE_AI_API_KEY_KEY) || "")
+            setModelId(localStorage.getItem(STORAGE_AI_MODEL_KEY) || "")
+
+            setError("")
         }
     }, [open])
 
-    const handleSave = () => {
-        localStorage.setItem(STORAGE_ACCESS_CODE_KEY, accessCode.trim())
-        onOpenChange(false)
+    const handleSave = async () => {
+        if (!accessCodeRequired) return
+
+        setError("")
+        setIsVerifying(true)
+
+        try {
+            const response = await fetch("/api/verify-access-code", {
+                method: "POST",
+                headers: {
+                    "x-access-code": accessCode.trim(),
+                },
+            })
+
+            const data = await response.json()
+
+            if (!data.valid) {
+                setError(data.message || "Invalid access code")
+                return
+            }
+
+            localStorage.setItem(STORAGE_ACCESS_CODE_KEY, accessCode.trim())
+            onOpenChange(false)
+        } catch {
+            setError("Failed to verify access code")
+        } finally {
+            setIsVerifying(false)
+        }
     }
 
     const handleKeyDown = (e: React.KeyboardEvent) => {
@@ -48,36 +155,281 @@ export function SettingsDialog({ open, onOpenChange }: SettingsDialogProps) {
                 <DialogHeader>
                     <DialogTitle>Settings</DialogTitle>
                     <DialogDescription>
-                        Configure your access settings.
+                        Configure your application settings.
                     </DialogDescription>
                 </DialogHeader>
                 <div className="space-y-4 py-2">
+                    {accessCodeRequired && (
+                        <div className="space-y-2">
+                            <Label htmlFor="access-code">Access Code</Label>
+                            <div className="flex gap-2">
+                                <Input
+                                    id="access-code"
+                                    type="password"
+                                    value={accessCode}
+                                    onChange={(e) =>
+                                        setAccessCode(e.target.value)
+                                    }
+                                    onKeyDown={handleKeyDown}
+                                    placeholder="Enter access code"
+                                    autoComplete="off"
+                                />
+                                <Button
+                                    onClick={handleSave}
+                                    disabled={isVerifying || !accessCode.trim()}
+                                >
+                                    {isVerifying ? "..." : "Save"}
+                                </Button>
+                            </div>
+                            <p className="text-[0.8rem] text-muted-foreground">
+                                Required to use this application.
+                            </p>
+                            {error && (
+                                <p className="text-[0.8rem] text-destructive">
+                                    {error}
+                                </p>
+                            )}
+                        </div>
+                    )}
                     <div className="space-y-2">
-                        <label className="text-sm font-medium leading-none peer-disabled:cursor-not-allowed peer-disabled:opacity-70">
-                            Access Code
-                        </label>
-                        <Input
-                            type="password"
-                            value={accessCode}
-                            onChange={(e) => setAccessCode(e.target.value)}
-                            onKeyDown={handleKeyDown}
-                            placeholder="Enter access code"
-                            autoComplete="off"
-                        />
+                        <Label>AI Provider Settings</Label>
                         <p className="text-[0.8rem] text-muted-foreground">
-                            Required if the server has enabled access control.
+                            Use your own API key to bypass usage limits. Your
+                            key is stored locally in your browser and is never
+                            stored on the server.
                         </p>
+                        <div className="space-y-3 pt-2">
+                            <div className="space-y-2">
+                                <Label htmlFor="ai-provider">Provider</Label>
+                                <Select
+                                    value={provider || "default"}
+                                    onValueChange={(value) => {
+                                        const actualValue =
+                                            value === "default" ? "" : value
+                                        setProvider(actualValue)
+                                        localStorage.setItem(
+                                            STORAGE_AI_PROVIDER_KEY,
+                                            actualValue,
+                                        )
+                                    }}
+                                >
+                                    <SelectTrigger id="ai-provider">
+                                        <SelectValue placeholder="Use Server Default" />
+                                    </SelectTrigger>
+                                    <SelectContent>
+                                        <SelectItem value="default">
+                                            Use Server Default
+                                        </SelectItem>
+                                        <SelectItem value="openai">
+                                            OpenAI
+                                        </SelectItem>
+                                        <SelectItem value="anthropic">
+                                            Anthropic
+                                        </SelectItem>
+                                        <SelectItem value="google">
+                                            Google
+                                        </SelectItem>
+                                        <SelectItem value="azure">
+                                            Azure OpenAI
+                                        </SelectItem>
+                                        <SelectItem value="openrouter">
+                                            OpenRouter
+                                        </SelectItem>
+                                        <SelectItem value="deepseek">
+                                            DeepSeek
+                                        </SelectItem>
+                                        <SelectItem value="siliconflow">
+                                            SiliconFlow
+                                        </SelectItem>
+                                    </SelectContent>
+                                </Select>
+                            </div>
+                            {provider && provider !== "default" && (
+                                <>
+                                    <div className="space-y-2">
+                                        <Label htmlFor="ai-model">
+                                            Model ID
+                                        </Label>
+                                        <Input
+                                            id="ai-model"
+                                            value={modelId}
+                                            onChange={(e) => {
+                                                setModelId(e.target.value)
+                                                localStorage.setItem(
+                                                    STORAGE_AI_MODEL_KEY,
+                                                    e.target.value,
+                                                )
+                                            }}
+                                            placeholder={
+                                                provider === "openai"
+                                                    ? "e.g., gpt-4o"
+                                                    : provider === "anthropic"
+                                                      ? "e.g., claude-sonnet-4-5"
+                                                      : provider === "google"
+                                                        ? "e.g., gemini-2.0-flash-exp"
+                                                        : provider ===
+                                                            "deepseek"
+                                                          ? "e.g., deepseek-chat"
+                                                          : "Model ID"
+                                            }
+                                        />
+                                    </div>
+                                    <div className="space-y-2">
+                                        <Label htmlFor="ai-api-key">
+                                            API Key
+                                        </Label>
+                                        <Input
+                                            id="ai-api-key"
+                                            type="password"
+                                            value={apiKey}
+                                            onChange={(e) => {
+                                                setApiKey(e.target.value)
+                                                localStorage.setItem(
+                                                    STORAGE_AI_API_KEY_KEY,
+                                                    e.target.value,
+                                                )
+                                            }}
+                                            placeholder="Your API key"
+                                            autoComplete="off"
+                                        />
+                                        <p className="text-[0.8rem] text-muted-foreground">
+                                            Overrides{" "}
+                                            {provider === "openai"
+                                                ? "OPENAI_API_KEY"
+                                                : provider === "anthropic"
+                                                  ? "ANTHROPIC_API_KEY"
+                                                  : provider === "google"
+                                                    ? "GOOGLE_GENERATIVE_AI_API_KEY"
+                                                    : provider === "azure"
+                                                      ? "AZURE_API_KEY"
+                                                      : provider ===
+                                                          "openrouter"
+                                                        ? "OPENROUTER_API_KEY"
+                                                        : provider ===
+                                                            "deepseek"
+                                                          ? "DEEPSEEK_API_KEY"
+                                                          : provider ===
+                                                              "siliconflow"
+                                                            ? "SILICONFLOW_API_KEY"
+                                                            : "server API key"}
+                                        </p>
+                                    </div>
+                                    <div className="space-y-2">
+                                        <Label htmlFor="ai-base-url">
+                                            Base URL (optional)
+                                        </Label>
+                                        <Input
+                                            id="ai-base-url"
+                                            value={baseUrl}
+                                            onChange={(e) => {
+                                                setBaseUrl(e.target.value)
+                                                localStorage.setItem(
+                                                    STORAGE_AI_BASE_URL_KEY,
+                                                    e.target.value,
+                                                )
+                                            }}
+                                            placeholder={
+                                                provider === "anthropic"
+                                                    ? "https://api.anthropic.com/v1"
+                                                    : provider === "siliconflow"
+                                                      ? "https://api.siliconflow.com/v1"
+                                                      : "Custom endpoint URL"
+                                            }
+                                        />
+                                    </div>
+                                    <Button
+                                        variant="outline"
+                                        size="sm"
+                                        className="w-full"
+                                        onClick={() => {
+                                            localStorage.removeItem(
+                                                STORAGE_AI_PROVIDER_KEY,
+                                            )
+                                            localStorage.removeItem(
+                                                STORAGE_AI_BASE_URL_KEY,
+                                            )
+                                            localStorage.removeItem(
+                                                STORAGE_AI_API_KEY_KEY,
+                                            )
+                                            localStorage.removeItem(
+                                                STORAGE_AI_MODEL_KEY,
+                                            )
+                                            setProvider("")
+                                            setBaseUrl("")
+                                            setApiKey("")
+                                            setModelId("")
+                                        }}
+                                    >
+                                        Clear Settings
+                                    </Button>
+                                </>
+                            )}
+                        </div>
+                    </div>
+
+                    <div className="flex items-center justify-between">
+                        <div className="space-y-0.5">
+                            <Label htmlFor="theme-toggle">Theme</Label>
+                            <p className="text-[0.8rem] text-muted-foreground">
+                                Dark/Light mode for interface and DrawIO canvas.
+                            </p>
+                        </div>
+                        <Button
+                            id="theme-toggle"
+                            variant="outline"
+                            size="icon"
+                            onClick={onToggleDarkMode}
+                        >
+                            {darkMode ? (
+                                <Sun className="h-4 w-4" />
+                            ) : (
+                                <Moon className="h-4 w-4" />
+                            )}
+                        </Button>
+                    </div>
+
+                    <div className="flex items-center justify-between">
+                        <div className="space-y-0.5">
+                            <Label htmlFor="drawio-ui">DrawIO Style</Label>
+                            <p className="text-[0.8rem] text-muted-foreground">
+                                Canvas style:{" "}
+                                {drawioUi === "min" ? "Minimal" : "Sketch"}
+                            </p>
+                        </div>
+                        <Button
+                            id="drawio-ui"
+                            variant="outline"
+                            size="sm"
+                            onClick={onToggleDrawioUi}
+                        >
+                            Switch to{" "}
+                            {drawioUi === "min" ? "Sketch" : "Minimal"}
+                        </Button>
+                    </div>
+
+                    <div className="flex items-center justify-between">
+                        <div className="space-y-0.5">
+                            <Label htmlFor="close-protection">
+                                Close Protection
+                            </Label>
+                            <p className="text-[0.8rem] text-muted-foreground">
+                                Show confirmation when leaving the page.
+                            </p>
+                        </div>
+                        <Switch
+                            id="close-protection"
+                            checked={closeProtection}
+                            onCheckedChange={(checked) => {
+                                setCloseProtection(checked)
+                                localStorage.setItem(
+                                    STORAGE_CLOSE_PROTECTION_KEY,
+                                    checked.toString(),
+                                )
+                                onCloseProtectionChange?.(checked)
+                            }}
+                        />
                     </div>
                 </div>
-                <DialogFooter>
-                    <Button
-                        variant="outline"
-                        onClick={() => onOpenChange(false)}
-                    >
-                        Cancel
-                    </Button>
-                    <Button onClick={handleSave}>Save</Button>
-                </DialogFooter>
             </DialogContent>
         </Dialog>
     )

components/ui/collapsible.tsx

@@ -0,0 +1,33 @@
+"use client"
+
+import * as CollapsiblePrimitive from "@radix-ui/react-collapsible"
+
+function Collapsible({
+  ...props
+}: React.ComponentProps<typeof CollapsiblePrimitive.Root>) {
+  return <CollapsiblePrimitive.Root data-slot="collapsible" {...props} />
+}
+
+function CollapsibleTrigger({
+  ...props
+}: React.ComponentProps<typeof CollapsiblePrimitive.CollapsibleTrigger>) {
+  return (
+    <CollapsiblePrimitive.CollapsibleTrigger
+      data-slot="collapsible-trigger"
+      {...props}
+    />
+  )
+}
+
+function CollapsibleContent({
+  ...props
+}: React.ComponentProps<typeof CollapsiblePrimitive.CollapsibleContent>) {
+  return (
+    <CollapsiblePrimitive.CollapsibleContent
+      data-slot="collapsible-content"
+      {...props}
+    />
+  )
+}
+
+export { Collapsible, CollapsibleTrigger, CollapsibleContent }

components/ui/label.tsx

@@ -0,0 +1,24 @@
+"use client"
+
+import * as React from "react"
+import * as LabelPrimitive from "@radix-ui/react-label"
+
+import { cn } from "@/lib/utils"
+
+function Label({
+  className,
+  ...props
+}: React.ComponentProps<typeof LabelPrimitive.Root>) {
+  return (
+    <LabelPrimitive.Root
+      data-slot="label"
+      className={cn(
+        "flex items-center gap-2 text-sm leading-none font-medium select-none group-data-[disabled=true]:pointer-events-none group-data-[disabled=true]:opacity-50 peer-disabled:cursor-not-allowed peer-disabled:opacity-50",
+        className
+      )}
+      {...props}
+    />
+  )
+}
+
+export { Label }

components/ui/switch.tsx

@@ -0,0 +1,31 @@
+"use client"
+
+import * as React from "react"
+import * as SwitchPrimitive from "@radix-ui/react-switch"
+
+import { cn } from "@/lib/utils"
+
+function Switch({
+  className,
+  ...props
+}: React.ComponentProps<typeof SwitchPrimitive.Root>) {
+  return (
+    <SwitchPrimitive.Root
+      data-slot="switch"
+      className={cn(
+        "peer data-[state=checked]:bg-primary data-[state=unchecked]:bg-input focus-visible:border-ring focus-visible:ring-ring/50 dark:data-[state=unchecked]:bg-input/80 inline-flex h-[1.15rem] w-8 shrink-0 items-center rounded-full border border-transparent shadow-xs transition-all outline-none focus-visible:ring-[3px] disabled:cursor-not-allowed disabled:opacity-50",
+        className
+      )}
+      {...props}
+    >
+      <SwitchPrimitive.Thumb
+        data-slot="switch-thumb"
+        className={cn(
+          "bg-background dark:data-[state=unchecked]:bg-foreground dark:data-[state=checked]:bg-primary-foreground pointer-events-none block size-4 rounded-full ring-0 transition-transform data-[state=checked]:translate-x-[calc(100%-2px)] data-[state=unchecked]:translate-x-0"
+        )}
+      />
+    </SwitchPrimitive.Root>
+  )
+}
+
+export { Switch }

contexts/diagram-context.tsx

@@ -3,14 +3,15 @@
 import type React from "react"
 import { createContext, useContext, useRef, useState } from "react"
 import type { DrawIoEmbedRef } from "react-drawio"
+import { STORAGE_DIAGRAM_XML_KEY } from "@/components/chat-panel"
 import type { ExportFormat } from "@/components/save-dialog"
-import { extractDiagramXML } from "../lib/utils"
+import { extractDiagramXML, validateAndFixXml } from "../lib/utils"
 
 interface DiagramContextType {
     chartXML: string
     latestSvg: string
     diagramHistory: { svg: string; xml: string }[]
-    loadDiagram: (chart: string) => void
+    loadDiagram: (chart: string, skipValidation?: boolean) => string | null
     handleExport: () => void
     handleExportWithoutHistory: () => void
     resolverRef: React.Ref<((value: string) => void) | null>
@@ -22,6 +23,10 @@ interface DiagramContextType {
         format: ExportFormat,
         sessionId?: string,
     ) => void
+    saveDiagramToStorage: () => Promise<void>
+    isDrawioReady: boolean
+    onDrawioLoad: () => void
+    resetDrawioReady: () => void
 }
 
 const DiagramContext = createContext<DiagramContextType | undefined>(undefined)
@@ -32,10 +37,27 @@ export function DiagramProvider({ children }: { children: React.ReactNode }) {
     const [diagramHistory, setDiagramHistory] = useState<
         { svg: string; xml: string }[]
     >([])
+    const [isDrawioReady, setIsDrawioReady] = useState(false)
+    const hasCalledOnLoadRef = useRef(false)
     const drawioRef = useRef<DrawIoEmbedRef | null>(null)
     const resolverRef = useRef<((value: string) => void) | null>(null)
     // Track if we're expecting an export for history (user-initiated)
     const expectHistoryExportRef = useRef<boolean>(false)
+
+    const onDrawioLoad = () => {
+        // Only set ready state once to prevent infinite loops
+        if (hasCalledOnLoadRef.current) return
+        hasCalledOnLoadRef.current = true
+        // console.log("[DiagramContext] DrawIO loaded, setting ready state")
+        setIsDrawioReady(true)
+    }
+
+    const resetDrawioReady = () => {
+        // console.log("[DiagramContext] Resetting DrawIO ready state")
+        hasCalledOnLoadRef.current = false
+        setIsDrawioReady(false)
+    }
+
     // Track if we're expecting an export for file save (stores raw export data)
     const saveResolverRef = useRef<{
         resolver: ((data: string) => void) | null
@@ -61,12 +83,66 @@ export function DiagramProvider({ children }: { children: React.ReactNode }) {
         }
     }
 
-    const loadDiagram = (chart: string) => {
+    // Save current diagram to localStorage (used before theme/UI changes)
+    const saveDiagramToStorage = async (): Promise<void> => {
+        if (!drawioRef.current) return
+
+        try {
+            const currentXml = await Promise.race([
+                new Promise<string>((resolve) => {
+                    resolverRef.current = resolve
+                    drawioRef.current?.exportDiagram({ format: "xmlsvg" })
+                }),
+                new Promise<string>((_, reject) =>
+                    setTimeout(() => reject(new Error("Export timeout")), 2000),
+                ),
+            ])
+
+            // Only save if diagram has meaningful content (not empty template)
+            if (currentXml && currentXml.length > 300) {
+                localStorage.setItem(STORAGE_DIAGRAM_XML_KEY, currentXml)
+            }
+        } catch (error) {
+            console.error("Failed to save diagram to storage:", error)
+        }
+    }
+
+    const loadDiagram = (
+        chart: string,
+        skipValidation?: boolean,
+    ): string | null => {
+        let xmlToLoad = chart
+
+        // Validate XML structure before loading (unless skipped for internal use)
+        if (!skipValidation) {
+            const validation = validateAndFixXml(chart)
+            if (!validation.valid) {
+                console.warn(
+                    "[loadDiagram] Validation error:",
+                    validation.error,
+                )
+                return validation.error
+            }
+            // Use fixed XML if auto-fix was applied
+            if (validation.fixed) {
+                console.log(
+                    "[loadDiagram] Auto-fixed XML issues:",
+                    validation.fixes,
+                )
+                xmlToLoad = validation.fixed
+            }
+        }
+
+        // Keep chartXML in sync even when diagrams are injected (e.g., display_diagram tool)
+        setChartXML(xmlToLoad)
+
         if (drawioRef.current) {
             drawioRef.current.load({
-                xml: chart,
+                xml: xmlToLoad,
             })
         }
+
+        return null
     }
 
     const handleDiagramExport = (data: any) => {
@@ -87,14 +163,20 @@ export function DiagramProvider({ children }: { children: React.ReactNode }) {
         setLatestSvg(data.data)
 
         // Only add to history if this was a user-initiated export
+        // Limit to 20 entries to prevent memory leaks during long sessions
+        const MAX_HISTORY_SIZE = 20
         if (expectHistoryExportRef.current) {
-            setDiagramHistory((prev) => [
-                ...prev,
-                {
-                    svg: data.data,
-                    xml: extractedXML,
-                },
-            ])
+            setDiagramHistory((prev) => {
+                const newHistory = [
+                    ...prev,
+                    {
+                        svg: data.data,
+                        xml: extractedXML,
+                    },
+                ]
+                // Keep only the last MAX_HISTORY_SIZE entries (circular buffer)
+                return newHistory.slice(-MAX_HISTORY_SIZE)
+            })
             expectHistoryExportRef.current = false
         }
 
@@ -106,8 +188,8 @@ export function DiagramProvider({ children }: { children: React.ReactNode }) {
 
     const clearDiagram = () => {
         const emptyDiagram = `<mxfile><diagram name="Page-1" id="page-1"><mxGraphModel><root><mxCell id="0"/><mxCell id="1" parent="0"/></root></mxGraphModel></diagram></mxfile>`
-        loadDiagram(emptyDiagram)
-        setChartXML(emptyDiagram)
+        // Skip validation for trusted internal template (loadDiagram also sets chartXML)
+        loadDiagram(emptyDiagram, true)
         setLatestSvg("")
         setDiagramHistory([])
     }
@@ -142,6 +224,9 @@ export function DiagramProvider({ children }: { children: React.ReactNode }) {
                     fileContent = xmlContent
                     mimeType = "application/xml"
                     extension = ".drawio"
+
+                    // Save to localStorage when user manually saves
+                    localStorage.setItem(STORAGE_DIAGRAM_XML_KEY, xmlContent)
                 } else if (format === "png") {
                     // PNG data comes as base64 data URL
                     fileContent = exportData
@@ -220,6 +305,10 @@ export function DiagramProvider({ children }: { children: React.ReactNode }) {
                 handleDiagramExport,
                 clearDiagram,
                 saveDiagramToFile,
+                saveDiagramToStorage,
+                isDrawioReady,
+                onDrawioLoad,
+                resetDrawioReady,
             }}
         >
             {children}

docker-compose.yml

@@ -0,0 +1,12 @@
+services:
+  drawio:
+    image: jgraph/drawio:latest
+    ports: ["8080:8080"]
+  next-ai-draw-io:
+    build:
+      context: .
+      args:
+        - NEXT_PUBLIC_DRAWIO_BASE_URL=http://localhost:8080
+    ports: ["3000:3000"]
+    env_file: .env
+    depends_on: [drawio]

docs/README_CN.md

@@ -4,14 +4,16 @@
 
 **AI驱动的图表创建工具 - 对话、绘制、可视化**
 
-[English](./README.md) | 中文 | [日本語](./README_JA.md)
+[English](../README.md) | 中文 | [日本語](./README_JA.md)
 
-[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
-[![Next.js](https://img.shields.io/badge/Next.js-15.x-black)](https://nextjs.org/)
-[![TypeScript](https://img.shields.io/badge/TypeScript-5.x-blue)](https://www.typescriptlang.org/)
+[![TrendShift](https://trendshift.io/api/badge/repositories/15449)](https://next-ai-drawio.jiang.jp/)
+
+[![License: Apache 2.0](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
+[![Next.js](https://img.shields.io/badge/Next.js-16.x-black)](https://nextjs.org/)
+[![React](https://img.shields.io/badge/React-19.x-61dafb)](https://react.dev/)
 [![Sponsor](https://img.shields.io/badge/Sponsor-❤-ea4aaa)](https://github.com/sponsors/DayuanJiang)
 
-[🚀 在线演示](https://next-ai-drawio.jiang.jp/)
+[![Live Demo](../public/live-demo-button.svg)](https://next-ai-drawio.jiang.jp/)
 
 </div>
 
@@ -19,16 +21,24 @@
 
 https://github.com/user-attachments/assets/b2eef5f3-b335-4e71-a755-dc2e80931979
 
-## 功能特性
+## 目录
+- [Next AI Draw.io](#next-ai-drawio)
+  - [目录](#目录)
+  - [示例](#示例)
+  - [功能特性](#功能特性)
+  - [MCP服务器（预览）](#mcp服务器预览)
+  - [快速开始](#快速开始)
+    - [在线试用](#在线试用)
+    - [使用Docker运行（推荐）](#使用docker运行推荐)
+    - [安装](#安装)
+  - [部署](#部署)
+  - [多提供商支持](#多提供商支持)
+  - [工作原理](#工作原理)
+  - [项目结构](#项目结构)
+  - [支持与联系](#支持与联系)
+  - [Star历史](#star历史)
 
--   **LLM驱动的图表创建**：利用大语言模型通过自然语言命令直接创建和操作draw.io图表
--   **基于图像的图表复制**：上传现有图表或图像，让AI自动复制和增强
--   **图表历史记录**：全面的版本控制，跟踪所有更改，允许您查看和恢复AI编辑前的图表版本
--   **交互式聊天界面**：与AI实时对话来完善您的图表
--   **AWS架构图支持**：专门支持生成AWS架构图
--   **动画连接器**：在图表元素之间创建动态动画连接器，实现更好的可视化效果
-
-## **示例**
+## 示例
 
 以下是一些示例提示词及其生成的图表：
 
@@ -38,67 +48,89 @@ https://github.com/user-attachments/assets/b2eef5f3-b335-4e71-a755-dc2e80931979
     <td colspan="2" valign="top" align="center">
       <strong>动画Transformer连接器</strong><br />
       <p><strong>提示词：</strong> 给我一个带有**动画连接器**的Transformer架构图。</p>
-      <img src="./public/animated_connectors.svg" alt="带动画连接器的Transformer架构" width="480" />
+      <img src="../public/animated_connectors.svg" alt="带动画连接器的Transformer架构" width="480" />
     </td>
   </tr>
   <tr>
     <td width="50%" valign="top">
       <strong>GCP架构图</strong><br />
       <p><strong>提示词：</strong> 使用**GCP图标**生成一个GCP架构图。在这个图中，用户连接到托管在实例上的前端。</p>
-      <img src="./public/gcp_demo.svg" alt="GCP架构图" width="480" />
+      <img src="../public/gcp_demo.svg" alt="GCP架构图" width="480" />
     </td>
     <td width="50%" valign="top">
       <strong>AWS架构图</strong><br />
       <p><strong>提示词：</strong> 使用**AWS图标**生成一个AWS架构图。在这个图中，用户连接到托管在实例上的前端。</p>
-      <img src="./public/aws_demo.svg" alt="AWS架构图" width="480" />
+      <img src="../public/aws_demo.svg" alt="AWS架构图" width="480" />
     </td>
   </tr>
   <tr>
     <td width="50%" valign="top">
       <strong>Azure架构图</strong><br />
       <p><strong>提示词：</strong> 使用**Azure图标**生成一个Azure架构图。在这个图中，用户连接到托管在实例上的前端。</p>
-      <img src="./public/azure_demo.svg" alt="Azure架构图" width="480" />
+      <img src="../public/azure_demo.svg" alt="Azure架构图" width="480" />
     </td>
     <td width="50%" valign="top">
       <strong>猫咪素描</strong><br />
       <p><strong>提示词：</strong> 给我画一只可爱的猫。</p>
-      <img src="./public/cat_demo.svg" alt="猫咪绘图" width="240" />
+      <img src="../public/cat_demo.svg" alt="猫咪绘图" width="240" />
     </td>
   </tr>
 </table>
 </div>
 
-## 工作原理
+## 功能特性
 
-本应用使用以下技术：
+-   **LLM驱动的图表创建**：利用大语言模型通过自然语言命令直接创建和操作draw.io图表
+-   **基于图像的图表复制**：上传现有图表或图像，让AI自动复制和增强
+-   **PDF和文本文件上传**：上传PDF文档和文本文件，提取内容并从现有文档生成图表
+-   **AI推理过程显示**：查看支持模型的AI思考过程（OpenAI o1/o3、Gemini、Claude等）
+-   **图表历史记录**：全面的版本控制，跟踪所有更改，允许您查看和恢复AI编辑前的图表版本
+-   **交互式聊天界面**：与AI实时对话来完善您的图表
+-   **云架构图支持**：专门支持生成云架构图（AWS、GCP、Azure）
+-   **动画连接器**：在图表元素之间创建动态动画连接器，实现更好的可视化效果
 
--   **Next.js**：用于前端框架和路由
--   **Vercel AI SDK**（`ai` + `@ai-sdk/*`）：用于流式AI响应和多提供商支持
--   **react-drawio**：用于图表表示和操作
+## MCP服务器（预览）
 
-图表以XML格式表示，可在draw.io中渲染。AI处理您的命令并相应地生成或修改此XML。
+> **预览功能**：此功能为实验性功能，可能会有变化。
 
-## 多提供商支持
+通过MCP（模型上下文协议）在Claude Desktop、Cursor和VS Code等AI代理中使用Next AI Draw.io。
 
--   AWS Bedrock（默认）
--   OpenAI
--   Anthropic
--   Google AI
--   Azure OpenAI
--   Ollama
--   OpenRouter
--   DeepSeek
+```json
+{
+  "mcpServers": {
+    "drawio": {
+      "command": "npx",
+      "args": ["@next-ai-drawio/mcp-server@latest"]
+    }
+  }
+}
+```
 
-除AWS Bedrock和OpenRouter外，所有提供商都支持自定义端点。
+### Claude Code CLI
 
-📖 **[详细的提供商配置指南](./docs/ai-providers.md)** - 查看各提供商的设置说明。
+```bash
+claude mcp add drawio -- npx @next-ai-drawio/mcp-server@latest
+```
 
-**模型要求**：此任务需要强大的模型能力，因为它涉及生成具有严格格式约束的长文本（draw.io XML）。推荐使用Claude Sonnet 4.5、GPT-4o、Gemini 2.0和DeepSeek V3/R1。
+然后让Claude创建图表：
+> "创建一个展示用户认证流程的流程图，包含登录、MFA和会话管理"
 
-注意：`claude-sonnet-4-5` 已在带有AWS标志的draw.io图表上进行训练，因此如果您想创建AWS架构图，这是最佳选择。
+图表会实时显示在浏览器中！
+
+详情请参阅[MCP服务器README](../packages/mcp-server/README.md)，了解VS Code、Cursor等客户端配置。
 
 ## 快速开始
 
+### 在线试用
+
+无需安装！直接在我们的演示站点试用：
+
+[![Live Demo](../public/live-demo-button.svg)](https://next-ai-drawio.jiang.jp/)
+
+> 注意：由于访问量较大，演示站点目前使用 minimax-m2 模型。如需获得最佳效果，建议使用 Claude Sonnet 4.5 或 Claude Opus 4.5 自行部署。
+
+> **使用自己的 API Key**：您可以使用自己的 API Key 来绕过演示站点的用量限制。点击聊天面板中的设置图标即可配置您的 Provider 和 API Key。您的 Key 仅保存在浏览器本地，不会被存储在服务器上。
+
 ### 使用Docker运行（推荐）
 
 如果您只想在本地运行，最好的方式是使用Docker。
@@ -115,10 +147,20 @@ docker run -d -p 3000:3000 \
   ghcr.io/dayuanjiang/next-ai-draw-io:latest
 ```
 
+或者使用 env 文件：
+
+```bash
+cp env.example .env
+# 编辑 .env 填写您的配置
+docker run -d -p 3000:3000 --env-file .env ghcr.io/dayuanjiang/next-ai-draw-io:latest
+```
+
 在浏览器中打开 [http://localhost:3000](http://localhost:3000)。
 
 请根据您首选的AI提供商配置替换环境变量。可用选项请参阅[多提供商支持](#多提供商支持)。
 
+> **离线部署：** 如果 `embed.diagrams.net` 被屏蔽，请参阅 [离线部署指南](./offline-deployment.md) 了解配置选项。
+
 ### 安装
 
 1. 克隆仓库：
@@ -132,8 +174,6 @@ cd next-ai-draw-io
 
 ```bash
 npm install
-# 或
-yarn install
 ```
 
 3. 配置您的AI提供商：
@@ -146,14 +186,15 @@ cp env.example .env.local
 
 编辑 `.env.local` 并配置您选择的提供商：
 
--   将 `AI_PROVIDER` 设置为您选择的提供商（bedrock, openai, anthropic, google, azure, ollama, openrouter, deepseek）
+-   将 `AI_PROVIDER` 设置为您选择的提供商（bedrock, openai, anthropic, google, azure, ollama, openrouter, deepseek, siliconflow）
 -   将 `AI_MODEL` 设置为您要使用的特定模型
 -   添加您的提供商所需的API密钥
+-   `TEMPERATURE`：可选的温度设置（例如 `0` 表示确定性输出）。对于不支持此参数的模型（如推理模型），请不要设置。
 -   `ACCESS_CODE_LIST` 访问密码，可选，可以使用逗号隔开多个密码。
 
 > 警告：如果不填写 `ACCESS_CODE_LIST`，则任何人都可以直接使用你部署后的网站，可能会导致你的 token 被急速消耗完毕，建议填写此选项。
 
-详细设置说明请参阅[提供商配置指南](./docs/ai-providers.md)。
+详细设置说明请参阅[提供商配置指南](./ai-providers.md)。
 
 4. 运行开发服务器：
 
@@ -174,6 +215,38 @@ npm run dev
 
 请确保在Vercel控制台中**设置环境变量**，就像您在本地 `.env.local` 文件中所做的那样。
 
+
+## 多提供商支持
+
+-   AWS Bedrock（默认）
+-   OpenAI
+-   Anthropic
+-   Google AI
+-   Azure OpenAI
+-   Ollama
+-   OpenRouter
+-   DeepSeek
+-   SiliconFlow
+
+除AWS Bedrock和OpenRouter外，所有提供商都支持自定义端点。
+
+📖 **[详细的提供商配置指南](./ai-providers.md)** - 查看各提供商的设置说明。
+
+**模型要求**：此任务需要强大的模型能力，因为它涉及生成具有严格格式约束的长文本（draw.io XML）。推荐使用Claude Sonnet 4.5、GPT-4o、Gemini 2.0和DeepSeek V3/R1。
+
+注意：`claude-sonnet-4-5` 已在带有AWS标志的draw.io图表上进行训练，因此如果您想创建AWS架构图，这是最佳选择。
+
+
+## 工作原理
+
+本应用使用以下技术：
+
+-   **Next.js**：用于前端框架和路由
+-   **Vercel AI SDK**（`ai` + `@ai-sdk/*`）：用于流式AI响应和多提供商支持
+-   **react-drawio**：用于图表表示和操作
+
+图表以XML格式表示，可在draw.io中渲染。AI处理您的命令并相应地生成或修改此XML。
+
 ## 项目结构
 
 ```
@@ -193,14 +266,6 @@ lib/                  # 工具函数和辅助程序
 public/               # 静态资源包括示例图片
 ```
 
-## 待办事项
-
--   [x] 允许LLM修改XML而不是每次从头生成
--   [x] 提高形状流式更新的流畅度
--   [x] 添加多AI提供商支持（OpenAI, Anthropic, Google, Azure, Ollama）
--   [x] 解决超过60秒的会话生成失败的bug
--   [ ] 在UI上添加API配置
-
 ## 支持与联系
 
 如果您觉得这个项目有用，请考虑[赞助](https://github.com/sponsors/DayuanJiang)来帮助我托管在线演示站点！

docs/README_JA.md

@@ -4,14 +4,16 @@
 
 **AI搭載のダイアグラム作成ツール - チャット、描画、可視化**
 
-[English](./README.md) | [中文](./README_CN.md) | 日本語
+[English](../README.md) | [中文](./README_CN.md) | 日本語
 
-[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
-[![Next.js](https://img.shields.io/badge/Next.js-15.x-black)](https://nextjs.org/)
-[![TypeScript](https://img.shields.io/badge/TypeScript-5.x-blue)](https://www.typescriptlang.org/)
+[![TrendShift](https://trendshift.io/api/badge/repositories/15449)](https://next-ai-drawio.jiang.jp/)
+
+[![License: Apache 2.0](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
+[![Next.js](https://img.shields.io/badge/Next.js-16.x-black)](https://nextjs.org/)
+[![React](https://img.shields.io/badge/React-19.x-61dafb)](https://react.dev/)
 [![Sponsor](https://img.shields.io/badge/Sponsor-❤-ea4aaa)](https://github.com/sponsors/DayuanJiang)
 
-[🚀 ライブデモ](https://next-ai-drawio.jiang.jp/)
+[![Live Demo](../public/live-demo-button.svg)](https://next-ai-drawio.jiang.jp/)
 
 </div>
 
@@ -19,16 +21,24 @@ AI機能とdraw.ioダイアグラムを統合したNext.jsウェブアプリケ
 
 https://github.com/user-attachments/assets/b2eef5f3-b335-4e71-a755-dc2e80931979
 
-## 機能
+## 目次
+- [Next AI Draw.io](#next-ai-drawio)
+  - [目次](#目次)
+  - [例](#例)
+  - [機能](#機能)
+  - [MCPサーバー（プレビュー）](#mcpサーバープレビュー)
+  - [はじめに](#はじめに)
+    - [オンラインで試す](#オンラインで試す)
+    - [Dockerで実行（推奨）](#dockerで実行推奨)
+    - [インストール](#インストール)
+  - [デプロイ](#デプロイ)
+  - [マルチプロバイダーサポート](#マルチプロバイダーサポート)
+  - [仕組み](#仕組み)
+  - [プロジェクト構造](#プロジェクト構造)
+  - [サポート＆お問い合わせ](#サポートお問い合わせ)
+  - [スター履歴](#スター履歴)
 
--   **LLM搭載のダイアグラム作成**：大規模言語モデルを活用して、自然言語コマンドで直接draw.ioダイアグラムを作成・操作
--   **画像ベースのダイアグラム複製**：既存のダイアグラムや画像をアップロードし、AIが自動的に複製・強化
--   **ダイアグラム履歴**：すべての変更を追跡する包括的なバージョン管理。AI編集前のダイアグラムの以前のバージョンを表示・復元可能
--   **インタラクティブなチャットインターフェース**：AIとリアルタイムでコミュニケーションしてダイアグラムを改善
--   **AWSアーキテクチャダイアグラムサポート**：AWSアーキテクチャダイアグラムの生成を専門的にサポート
--   **アニメーションコネクタ**：より良い可視化のためにダイアグラム要素間に動的でアニメーション化されたコネクタを作成
-
-## **例**
+## 例
 
 以下はいくつかのプロンプト例と生成されたダイアグラムです：
 
@@ -38,67 +48,89 @@ https://github.com/user-attachments/assets/b2eef5f3-b335-4e71-a755-dc2e80931979
     <td colspan="2" valign="top" align="center">
       <strong>アニメーションTransformerコネクタ</strong><br />
       <p><strong>プロンプト：</strong> **アニメーションコネクタ**付きのTransformerアーキテクチャ図を作成してください。</p>
-      <img src="./public/animated_connectors.svg" alt="アニメーションコネクタ付きTransformerアーキテクチャ" width="480" />
+      <img src="../public/animated_connectors.svg" alt="アニメーションコネクタ付きTransformerアーキテクチャ" width="480" />
     </td>
   </tr>
   <tr>
     <td width="50%" valign="top">
       <strong>GCPアーキテクチャ図</strong><br />
       <p><strong>プロンプト：</strong> **GCPアイコン**を使用してGCPアーキテクチャ図を生成してください。この図では、ユーザーがインスタンス上でホストされているフロントエンドに接続します。</p>
-      <img src="./public/gcp_demo.svg" alt="GCPアーキテクチャ図" width="480" />
+      <img src="../public/gcp_demo.svg" alt="GCPアーキテクチャ図" width="480" />
     </td>
     <td width="50%" valign="top">
       <strong>AWSアーキテクチャ図</strong><br />
       <p><strong>プロンプト：</strong> **AWSアイコン**を使用してAWSアーキテクチャ図を生成してください。この図では、ユーザーがインスタンス上でホストされているフロントエンドに接続します。</p>
-      <img src="./public/aws_demo.svg" alt="AWSアーキテクチャ図" width="480" />
+      <img src="../public/aws_demo.svg" alt="AWSアーキテクチャ図" width="480" />
     </td>
   </tr>
   <tr>
     <td width="50%" valign="top">
       <strong>Azureアーキテクチャ図</strong><br />
       <p><strong>プロンプト：</strong> **Azureアイコン**を使用してAzureアーキテクチャ図を生成してください。この図では、ユーザーがインスタンス上でホストされているフロントエンドに接続します。</p>
-      <img src="./public/azure_demo.svg" alt="Azureアーキテクチャ図" width="480" />
+      <img src="../public/azure_demo.svg" alt="Azureアーキテクチャ図" width="480" />
     </td>
     <td width="50%" valign="top">
       <strong>猫のスケッチ</strong><br />
       <p><strong>プロンプト：</strong> かわいい猫を描いてください。</p>
-      <img src="./public/cat_demo.svg" alt="猫の絵" width="240" />
+      <img src="../public/cat_demo.svg" alt="猫の絵" width="240" />
     </td>
   </tr>
 </table>
 </div>
 
-## 仕組み
+## 機能
 
-本アプリケーションは以下の技術を使用しています：
+-   **LLM搭載のダイアグラム作成**：大規模言語モデルを活用して、自然言語コマンドで直接draw.ioダイアグラムを作成・操作
+-   **画像ベースのダイアグラム複製**：既存のダイアグラムや画像をアップロードし、AIが自動的に複製・強化
+-   **PDFとテキストファイルのアップロード**：PDFドキュメントやテキストファイルをアップロードして、既存のドキュメントからコンテンツを抽出し、ダイアグラムを生成
+-   **AI推論プロセス表示**：サポートされているモデル（OpenAI o1/o3、Gemini、Claudeなど）のAIの思考プロセスを表示
+-   **ダイアグラム履歴**：すべての変更を追跡する包括的なバージョン管理。AI編集前のダイアグラムの以前のバージョンを表示・復元可能
+-   **インタラクティブなチャットインターフェース**：AIとリアルタイムでコミュニケーションしてダイアグラムを改善
+-   **クラウドアーキテクチャダイアグラムサポート**：クラウドアーキテクチャダイアグラムの生成を専門的にサポート（AWS、GCP、Azure）
+-   **アニメーションコネクタ**：より良い可視化のためにダイアグラム要素間に動的でアニメーション化されたコネクタを作成
 
--   **Next.js**：フロントエンドフレームワークとルーティング
--   **Vercel AI SDK**（`ai` + `@ai-sdk/*`）：ストリーミングAIレスポンスとマルチプロバイダーサポート
--   **react-drawio**：ダイアグラムの表現と操作
+## MCPサーバー（プレビュー）
 
-ダイアグラムはdraw.ioでレンダリングできるXMLとして表現されます。AIがコマンドを処理し、それに応じてこのXMLを生成または変更します。
+> **プレビュー機能**：この機能は実験的であり、変更される可能性があります。
 
-## マルチプロバイダーサポート
+MCP（Model Context Protocol）を介して、Claude Desktop、Cursor、VS CodeなどのAIエージェントでNext AI Draw.ioを使用できます。
 
--   AWS Bedrock（デフォルト）
--   OpenAI
--   Anthropic
--   Google AI
--   Azure OpenAI
--   Ollama
--   OpenRouter
--   DeepSeek
+```json
+{
+  "mcpServers": {
+    "drawio": {
+      "command": "npx",
+      "args": ["@next-ai-drawio/mcp-server@latest"]
+    }
+  }
+}
+```
 
-AWS BedrockとOpenRouter以外のすべてのプロバイダーはカスタムエンドポイントをサポートしています。
+### Claude Code CLI
 
-📖 **[詳細なプロバイダー設定ガイド](./docs/ai-providers.md)** - 各プロバイダーの設定手順をご覧ください。
+```bash
+claude mcp add drawio -- npx @next-ai-drawio/mcp-server@latest
+```
 
-**モデル要件**：このタスクは厳密なフォーマット制約（draw.io XML）を持つ長文テキスト生成を伴うため、強力なモデル機能が必要です。Claude Sonnet 4.5、GPT-4o、Gemini 2.0、DeepSeek V3/R1を推奨します。
+Claudeにダイアグラムの作成を依頼：
+> 「ログイン、MFA、セッション管理を含むユーザー認証のフローチャートを作成してください」
 
-注：`claude-sonnet-4-5`はAWSロゴ付きのdraw.ioダイアグラムで学習されているため、AWSアーキテクチャダイアグラムを作成したい場合は最適な選択です。
+ダイアグラムがリアルタイムでブラウザに表示されます！
+
+詳細は[MCPサーバーREADME](../packages/mcp-server/README.md)をご覧ください（VS Code、Cursorなどのクライアント設定も含む）。
 
 ## はじめに
 
+### オンラインで試す
+
+インストール不要！デモサイトで直接お試しください：
+
+[![Live Demo](../public/live-demo-button.svg)](https://next-ai-drawio.jiang.jp/)
+
+> 注意：アクセス数が多いため、デモサイトでは現在 minimax-m2 モデルを使用しています。最高の結果を得るには、Claude Sonnet 4.5 または Claude Opus 4.5 でのセルフホスティングをお勧めします。
+
+> **自分のAPIキーを使用**：自分のAPIキーを使用することで、デモサイトの利用制限を回避できます。チャットパネルの設定アイコンをクリックして、プロバイダーとAPIキーを設定してください。キーはブラウザのローカルに保存され、サーバーには保存されません。
+
 ### Dockerで実行（推奨）
 
 ローカルで実行したいだけなら、Dockerを使用するのが最も簡単です。
@@ -115,10 +147,20 @@ docker run -d -p 3000:3000 \
   ghcr.io/dayuanjiang/next-ai-draw-io:latest
 ```
 
+または env ファイルを使用：
+
+```bash
+cp env.example .env
+# .env を編集して設定を入力
+docker run -d -p 3000:3000 --env-file .env ghcr.io/dayuanjiang/next-ai-draw-io:latest
+```
+
 ブラウザで [http://localhost:3000](http://localhost:3000) を開いてください。
 
 環境変数はお好みのAIプロバイダー設定に置き換えてください。利用可能なオプションについては[マルチプロバイダーサポート](#マルチプロバイダーサポート)を参照してください。
 
+> **オフラインデプロイ：** `embed.diagrams.net` がブロックされている場合は、[オフラインデプロイガイド](./offline-deployment.md) で設定オプションをご確認ください。
+
 ### インストール
 
 1. リポジトリをクローン：
@@ -132,8 +174,6 @@ cd next-ai-draw-io
 
 ```bash
 npm install
-# または
-yarn install
 ```
 
 3. AIプロバイダーを設定：
@@ -146,14 +186,15 @@ cp env.example .env.local
 
 `.env.local`を編集して選択したプロバイダーを設定：
 
--   `AI_PROVIDER`を選択したプロバイダーに設定（bedrock, openai, anthropic, google, azure, ollama, openrouter, deepseek）
+-   `AI_PROVIDER`を選択したプロバイダーに設定（bedrock, openai, anthropic, google, azure, ollama, openrouter, deepseek, siliconflow）
 -   `AI_MODEL`を使用する特定のモデルに設定
 -   プロバイダーに必要なAPIキーを追加
+-   `TEMPERATURE`：オプションの温度設定（例：`0`で決定論的な出力）。温度をサポートしないモデル（推論モデルなど）では設定しないでください。
 -   `ACCESS_CODE_LIST` アクセスパスワード（オプション）。カンマ区切りで複数のパスワードを指定できます。
 
 > 警告：`ACCESS_CODE_LIST`を設定しない場合、誰でもデプロイされたサイトに直接アクセスできるため、トークンが急速に消費される可能性があります。このオプションを設定することをお勧めします。
 
-詳細な設定手順については[プロバイダー設定ガイド](./docs/ai-providers.md)を参照してください。
+詳細な設定手順については[プロバイダー設定ガイド](./ai-providers.md)を参照してください。
 
 4. 開発サーバーを起動：
 
@@ -174,6 +215,38 @@ Next.jsアプリをデプロイする最も簡単な方法は、Next.jsの作成
 
 ローカルの`.env.local`ファイルと同様に、Vercelダッシュボードで**環境変数を設定**してください。
 
+
+## マルチプロバイダーサポート
+
+-   AWS Bedrock（デフォルト）
+-   OpenAI
+-   Anthropic
+-   Google AI
+-   Azure OpenAI
+-   Ollama
+-   OpenRouter
+-   DeepSeek
+-   SiliconFlow
+
+AWS BedrockとOpenRouter以外のすべてのプロバイダーはカスタムエンドポイントをサポートしています。
+
+📖 **[詳細なプロバイダー設定ガイド](./ai-providers.md)** - 各プロバイダーの設定手順をご覧ください。
+
+**モデル要件**：このタスクは厳密なフォーマット制約（draw.io XML）を持つ長文テキスト生成を伴うため、強力なモデル機能が必要です。Claude Sonnet 4.5、GPT-4o、Gemini 2.0、DeepSeek V3/R1を推奨します。
+
+注：`claude-sonnet-4-5`はAWSロゴ付きのdraw.ioダイアグラムで学習されているため、AWSアーキテクチャダイアグラムを作成したい場合は最適な選択です。
+
+
+## 仕組み
+
+本アプリケーションは以下の技術を使用しています：
+
+-   **Next.js**：フロントエンドフレームワークとルーティング
+-   **Vercel AI SDK**（`ai` + `@ai-sdk/*`）：ストリーミングAIレスポンスとマルチプロバイダーサポート
+-   **react-drawio**：ダイアグラムの表現と操作
+
+ダイアグラムはdraw.ioでレンダリングできるXMLとして表現されます。AIがコマンドを処理し、それに応じてこのXMLを生成または変更します。
+
 ## プロジェクト構造
 
 ```
@@ -193,14 +266,6 @@ lib/                  # ユーティリティ関数とヘルパー
 public/               # サンプル画像を含む静的アセット
 ```
 
-## TODO
-
--   [x] LLMが毎回ゼロから生成する代わりにXMLを修正できるようにする
--   [x] シェイプストリーミング更新の滑らかさを改善
--   [x] 複数のAIプロバイダーサポートを追加（OpenAI, Anthropic, Google, Azure, Ollama）
--   [x] 60秒以上のセッションで生成が失敗するバグを解決
--   [ ] UIにAPI設定を追加
-
 ## サポート＆お問い合わせ
 
 このプロジェクトが役に立ったら、ライブデモサイトのホスティングを支援するために[スポンサー](https://github.com/sponsors/DayuanJiang)をご検討ください！

docs/ai-providers.md

@@ -63,17 +63,40 @@ Optional custom endpoint:
 DEEPSEEK_BASE_URL=https://your-custom-endpoint
 ```
 
+### SiliconFlow (OpenAI-compatible)
+
+```bash
+SILICONFLOW_API_KEY=your_api_key
+AI_MODEL=deepseek-ai/DeepSeek-V3  # example; use any SiliconFlow model id
+```
+
+Optional custom endpoint (defaults to the recommended domain):
+
+```bash
+SILICONFLOW_BASE_URL=https://api.siliconflow.com/v1  # or https://api.siliconflow.cn/v1
+```
+
 ### Azure OpenAI
 
 ```bash
 AZURE_API_KEY=your_api_key
+AZURE_RESOURCE_NAME=your-resource-name  # Required: your Azure resource name
 AI_MODEL=your-deployment-name
 ```
 
-Optional custom endpoint:
+Or use a custom endpoint instead of resource name:
 
 ```bash
-AZURE_BASE_URL=https://your-resource.openai.azure.com
+AZURE_API_KEY=your_api_key
+AZURE_BASE_URL=https://your-resource.openai.azure.com  # Alternative to AZURE_RESOURCE_NAME
+AI_MODEL=your-deployment-name
+```
+
+Optional reasoning configuration:
+
+```bash
+AZURE_REASONING_EFFORT=low      # Optional: low, medium, high
+AZURE_REASONING_SUMMARY=detailed  # Optional: none, brief, detailed
 ```
 
 ### AWS Bedrock
@@ -85,7 +108,7 @@ AWS_SECRET_ACCESS_KEY=your_secret_access_key
 AI_MODEL=anthropic.claude-sonnet-4-5-20250514-v1:0
 ```
 
-Note: On AWS (Amplify, Lambda, EC2 with IAM role), credentials are automatically obtained from the IAM role.
+Note: On AWS (Lambda, EC2 with IAM role), credentials are automatically obtained from the IAM role.
 
 ### OpenRouter
 
@@ -113,6 +136,23 @@ Optional custom URL:
 OLLAMA_BASE_URL=http://localhost:11434
 ```
 
+### Vercel AI Gateway
+
+Vercel AI Gateway provides unified access to multiple AI providers through a single API key. This simplifies authentication and allows you to switch between providers without managing multiple API keys.
+
+```bash
+AI_GATEWAY_API_KEY=your_gateway_api_key
+AI_MODEL=openai/gpt-4o
+```
+
+Model format uses `provider/model` syntax:
+
+-   `openai/gpt-4o` - OpenAI GPT-4o
+-   `anthropic/claude-sonnet-4-5` - Anthropic Claude Sonnet 4.5
+-   `google/gemini-2.0-flash` - Google Gemini 2.0 Flash
+
+Get your API key from the [Vercel AI Gateway dashboard](https://vercel.com/ai-gateway).
+
 ## Auto-Detection
 
 If you only configure **one** provider's API key, the system will automatically detect and use that provider. No need to set `AI_PROVIDER`.
@@ -120,7 +160,7 @@ If you only configure **one** provider's API key, the system will automatically
 If you configure **multiple** API keys, you must explicitly set `AI_PROVIDER`:
 
 ```bash
-AI_PROVIDER=google  # or: openai, anthropic, deepseek, azure, bedrock, openrouter, ollama
+AI_PROVIDER=google  # or: openai, anthropic, deepseek, siliconflow, azure, bedrock, openrouter, ollama, gateway
 ```
 
 ## Model Capability Requirements
@@ -133,6 +173,20 @@ This task requires exceptionally strong model capabilities, as it involves gener
 
 **Note on Ollama**: While Ollama is supported as a provider, it's generally not practical for this use case unless you're running high-capability models like DeepSeek R1 or Qwen3-235B locally.
 
+## Temperature Setting
+
+You can optionally configure the temperature via environment variable:
+
+```bash
+TEMPERATURE=0  # More deterministic output (recommended for diagrams)
+```
+
+**Important**: Leave `TEMPERATURE` unset for models that don't support temperature settings, such as:
+- GPT-5.1 and other reasoning models
+- Some specialized models
+
+When unset, the model uses its default behavior.
+
 ## Recommendations
 
 -   **Best experience**: Use models with vision support (GPT-4o, Claude, Gemini) for image-to-diagram features

docs/offline-deployment.md

@@ -0,0 +1,39 @@
+# Offline Deployment
+
+Deploy Next AI Draw.io offline by self-hosting draw.io to replace `embed.diagrams.net`.
+
+**Note:** `NEXT_PUBLIC_DRAWIO_BASE_URL` is a **build-time** variable. Changing it requires rebuilding the Docker image.
+
+## Docker Compose Setup
+
+1. Clone the repository and define API keys in `.env`.
+2. Create `docker-compose.yml`:
+
+```yaml
+services:
+  drawio:
+    image: jgraph/drawio:latest
+    ports: ["8080:8080"]
+  next-ai-draw-io:
+    build:
+      context: .
+      args:
+        - NEXT_PUBLIC_DRAWIO_BASE_URL=http://localhost:8080
+    ports: ["3000:3000"]
+    env_file: .env
+    depends_on: [drawio]
+```
+
+3. Run `docker compose up -d` and open `http://localhost:3000`.
+
+## Configuration & Critical Warning
+
+**The `NEXT_PUBLIC_DRAWIO_BASE_URL` must be accessible from the user's browser.**
+
+| Scenario | URL Value |
+|----------|-----------|
+| Localhost | `http://localhost:8080` |
+| Remote/Server | `http://YOUR_SERVER_IP:8080` or `https://drawio.your-domain.com` |
+
+**Do NOT use** internal Docker aliases like `http://drawio:8080`; the browser cannot resolve them.
+

env.example

@@ -1,6 +1,6 @@
 # AI Provider Configuration
 # AI_PROVIDER: Which provider to use
-# Options: bedrock, openai, anthropic, google, azure, ollama, openrouter, deepseek
+# Options: bedrock, openai, anthropic, google, azure, ollama, openrouter, deepseek, siliconflow, gateway
 # Default: bedrock
 AI_PROVIDER=bedrock
 
@@ -11,28 +11,49 @@ AI_MODEL=global.anthropic.claude-sonnet-4-5-20250929-v1:0
 # AWS_REGION=us-east-1
 # AWS_ACCESS_KEY_ID=your-access-key-id
 # AWS_SECRET_ACCESS_KEY=your-secret-access-key
+# Note: Claude and Nova models support reasoning/extended thinking
+# BEDROCK_REASONING_BUDGET_TOKENS=12000  # Optional: Claude reasoning budget in tokens (1024-64000)
+# BEDROCK_REASONING_EFFORT=medium        # Optional: Nova reasoning effort (low/medium/high)
 
 # OpenAI Configuration
 # OPENAI_API_KEY=sk-...
 # OPENAI_BASE_URL=https://api.openai.com/v1  # Optional: Custom OpenAI-compatible endpoint
 # OPENAI_ORGANIZATION=org-...  # Optional
 # OPENAI_PROJECT=proj_...      # Optional
+# Note: o1/o3/gpt-5 models automatically enable reasoning summary (default: detailed)
+# OPENAI_REASONING_EFFORT=low   # Optional: Reasoning effort (minimal/low/medium/high) - for o1/o3/gpt-5
+# OPENAI_REASONING_SUMMARY=detailed  # Optional: Override reasoning summary (none/brief/detailed)
 
 # Anthropic (Direct) Configuration
 # ANTHROPIC_API_KEY=sk-ant-...
 # ANTHROPIC_BASE_URL=https://your-custom-anthropic/v1
+# ANTHROPIC_THINKING_TYPE=enabled            # Optional: Anthropic extended thinking (enabled)
+# ANTHROPIC_THINKING_BUDGET_TOKENS=12000     # Optional: Budget for extended thinking in tokens
 
 # Google Generative AI Configuration
 # GOOGLE_GENERATIVE_AI_API_KEY=...
 # GOOGLE_BASE_URL=https://generativelanguage.googleapis.com/v1beta  # Optional: Custom endpoint
+# GOOGLE_CANDIDATE_COUNT=1                   # Optional: Number of candidates to generate
+# GOOGLE_TOP_K=40                            # Optional: Top K sampling parameter
+# GOOGLE_TOP_P=0.95                          # Optional: Nucleus sampling parameter
+# Note: Gemini 2.5/3 models automatically enable reasoning display (includeThoughts: true)
+# GOOGLE_THINKING_BUDGET=8192                # Optional: Gemini 2.5 thinking budget in tokens (for more/less thinking)
+# GOOGLE_THINKING_LEVEL=high                 # Optional: Gemini 3 thinking level (low/high)
 
 # Azure OpenAI Configuration
+# Configure endpoint using ONE of these methods:
+#   1. AZURE_RESOURCE_NAME - SDK constructs: https://{name}.openai.azure.com/openai/v1{path}
+#   2. AZURE_BASE_URL - SDK appends /v1{path} to your URL
+# If both are set, AZURE_BASE_URL takes precedence.
 # AZURE_RESOURCE_NAME=your-resource-name
 # AZURE_API_KEY=...
-# AZURE_BASE_URL=https://your-resource.openai.azure.com  # Optional: Custom endpoint (overrides resourceName)
+# AZURE_BASE_URL=https://your-resource.openai.azure.com/openai  # Alternative: Custom endpoint
+# AZURE_REASONING_EFFORT=low                 # Optional: Azure reasoning effort (low, medium, high)
+# AZURE_REASONING_SUMMARY=detailed
 
 # Ollama (Local) Configuration
 # OLLAMA_BASE_URL=http://localhost:11434/api  # Optional, defaults to localhost
+# OLLAMA_ENABLE_THINKING=true                 # Optional: Enable thinking for models that support it (e.g., qwen3)
 
 # OpenRouter Configuration
 # OPENROUTER_API_KEY=sk-or-v1-...
@@ -42,11 +63,36 @@ AI_MODEL=global.anthropic.claude-sonnet-4-5-20250929-v1:0
 # DEEPSEEK_API_KEY=sk-...
 # DEEPSEEK_BASE_URL=https://api.deepseek.com/v1  # Optional: Custom endpoint
 
+# SiliconFlow Configuration (OpenAI-compatible)
+# Base domain can be .com or .cn, defaults to https://api.siliconflow.com/v1
+# SILICONFLOW_API_KEY=sk-...
+# SILICONFLOW_BASE_URL=https://api.siliconflow.com/v1  # Optional: switch to https://api.siliconflow.cn/v1 if needed
+
+# Vercel AI Gateway Configuration
+# Get your API key from: https://vercel.com/ai-gateway
+# Model format: "provider/model" e.g., "openai/gpt-4o", "anthropic/claude-sonnet-4-5"
+# AI_GATEWAY_API_KEY=...
+
 # Langfuse Observability (Optional)
 # Enable LLM tracing and analytics - https://langfuse.com
 # LANGFUSE_PUBLIC_KEY=pk-lf-...
 # LANGFUSE_SECRET_KEY=sk-lf-...
 # LANGFUSE_BASEURL=https://cloud.langfuse.com  # EU region, use https://us.cloud.langfuse.com for US
 
+# Temperature (Optional)
+# Controls randomness in AI responses. Lower = more deterministic.
+# Leave unset for models that don't support temperature (e.g., GPT-5.1 reasoning models)
+# TEMPERATURE=0
+
 # Access Control (Optional)
 # ACCESS_CODE_LIST=your-secret-code,another-code
+
+# Draw.io Configuration (Optional)
+# NEXT_PUBLIC_DRAWIO_BASE_URL=https://embed.diagrams.net  # Default: https://embed.diagrams.net
+# Use this to point to a self-hosted draw.io instance
+
+# PDF Input Feature (Optional)
+# Enable PDF file upload to extract text and generate diagrams
+# Enabled by default. Set to "false" to disable.
+# ENABLE_PDF_INPUT=true
+# NEXT_PUBLIC_MAX_EXTRACTED_CHARS=150000  # Max characters for PDF/text extraction (default: 150000)

lib/ai-config.ts

@@ -0,0 +1,26 @@
+import { STORAGE_KEYS } from "./storage"
+
+/**
+ * Get AI configuration from localStorage.
+ * Returns API keys and settings for custom AI providers.
+ * Used to override server defaults when user provides their own API key.
+ */
+export function getAIConfig() {
+    if (typeof window === "undefined") {
+        return {
+            accessCode: "",
+            aiProvider: "",
+            aiBaseUrl: "",
+            aiApiKey: "",
+            aiModel: "",
+        }
+    }
+
+    return {
+        accessCode: localStorage.getItem(STORAGE_KEYS.accessCode) || "",
+        aiProvider: localStorage.getItem(STORAGE_KEYS.aiProvider) || "",
+        aiBaseUrl: localStorage.getItem(STORAGE_KEYS.aiBaseUrl) || "",
+        aiApiKey: localStorage.getItem(STORAGE_KEYS.aiApiKey) || "",
+        aiModel: localStorage.getItem(STORAGE_KEYS.aiModel) || "",
+    }
+}

lib/ai-providers.ts

@@ -2,6 +2,7 @@ import { createAmazonBedrock } from "@ai-sdk/amazon-bedrock"
 import { createAnthropic } from "@ai-sdk/anthropic"
 import { azure, createAzure } from "@ai-sdk/azure"
 import { createDeepSeek, deepseek } from "@ai-sdk/deepseek"
+import { gateway } from "@ai-sdk/gateway"
 import { createGoogleGenerativeAI, google } from "@ai-sdk/google"
 import { createOpenAI, openai } from "@ai-sdk/openai"
 import { fromNodeProviderChain } from "@aws-sdk/credential-providers"
@@ -17,6 +18,8 @@ export type ProviderName =
     | "ollama"
     | "openrouter"
     | "deepseek"
+    | "siliconflow"
+    | "gateway"
 
 interface ModelConfig {
     model: any
@@ -25,6 +28,25 @@ interface ModelConfig {
     modelId: string
 }
 
+export interface ClientOverrides {
+    provider?: string | null
+    baseUrl?: string | null
+    apiKey?: string | null
+    modelId?: string | null
+}
+
+// Providers that can be used with client-provided API keys
+const ALLOWED_CLIENT_PROVIDERS: ProviderName[] = [
+    "openai",
+    "anthropic",
+    "google",
+    "azure",
+    "openrouter",
+    "deepseek",
+    "siliconflow",
+    "gateway",
+]
+
 // Bedrock provider options for Anthropic beta features
 const BEDROCK_ANTHROPIC_BETA = {
     bedrock: {
@@ -37,6 +59,297 @@ const ANTHROPIC_BETA_HEADERS = {
     "anthropic-beta": "fine-grained-tool-streaming-2025-05-14",
 }
 
+/**
+ * Safely parse integer from environment variable with validation
+ */
+function parseIntSafe(
+    value: string | undefined,
+    varName: string,
+    min?: number,
+    max?: number,
+): number | undefined {
+    if (!value) return undefined
+    const parsed = Number.parseInt(value, 10)
+    if (Number.isNaN(parsed)) {
+        throw new Error(`${varName} must be a valid integer, got: ${value}`)
+    }
+    if (min !== undefined && parsed < min) {
+        throw new Error(`${varName} must be >= ${min}, got: ${parsed}`)
+    }
+    if (max !== undefined && parsed > max) {
+        throw new Error(`${varName} must be <= ${max}, got: ${parsed}`)
+    }
+    return parsed
+}
+
+/**
+ * Build provider-specific options from environment variables
+ * Supports various AI SDK providers with their unique configuration options
+ *
+ * Environment variables:
+ * - OPENAI_REASONING_EFFORT: OpenAI reasoning effort level (minimal/low/medium/high) - for o1/o3/gpt-5
+ * - OPENAI_REASONING_SUMMARY: OpenAI reasoning summary (none/brief/detailed) - auto-enabled for o1/o3/gpt-5
+ * - ANTHROPIC_THINKING_BUDGET_TOKENS: Anthropic thinking budget in tokens (1024-64000)
+ * - ANTHROPIC_THINKING_TYPE: Anthropic thinking type (enabled)
+ * - GOOGLE_THINKING_BUDGET: Google Gemini 2.5 thinking budget in tokens (1024-100000)
+ * - GOOGLE_THINKING_LEVEL: Google Gemini 3 thinking level (low/high)
+ * - AZURE_REASONING_EFFORT: Azure/OpenAI reasoning effort (low/medium/high)
+ * - AZURE_REASONING_SUMMARY: Azure reasoning summary (none/brief/detailed)
+ * - BEDROCK_REASONING_BUDGET_TOKENS: Bedrock Claude reasoning budget in tokens (1024-64000)
+ * - BEDROCK_REASONING_EFFORT: Bedrock Nova reasoning effort (low/medium/high)
+ * - OLLAMA_ENABLE_THINKING: Enable Ollama thinking mode (set to "true")
+ */
+function buildProviderOptions(
+    provider: ProviderName,
+    modelId?: string,
+): Record<string, any> | undefined {
+    const options: Record<string, any> = {}
+
+    switch (provider) {
+        case "openai": {
+            const reasoningEffort = process.env.OPENAI_REASONING_EFFORT
+            const reasoningSummary = process.env.OPENAI_REASONING_SUMMARY
+
+            // OpenAI reasoning models (o1, o3, gpt-5) need reasoningSummary to return thoughts
+            if (
+                modelId &&
+                (modelId.includes("o1") ||
+                    modelId.includes("o3") ||
+                    modelId.includes("gpt-5"))
+            ) {
+                options.openai = {
+                    // Auto-enable reasoning summary for reasoning models (default: detailed)
+                    reasoningSummary:
+                        (reasoningSummary as "none" | "brief" | "detailed") ||
+                        "detailed",
+                }
+
+                // Optionally configure reasoning effort
+                if (reasoningEffort) {
+                    options.openai.reasoningEffort = reasoningEffort as
+                        | "minimal"
+                        | "low"
+                        | "medium"
+                        | "high"
+                }
+            } else if (reasoningEffort || reasoningSummary) {
+                // Non-reasoning models: only apply if explicitly configured
+                options.openai = {}
+                if (reasoningEffort) {
+                    options.openai.reasoningEffort = reasoningEffort as
+                        | "minimal"
+                        | "low"
+                        | "medium"
+                        | "high"
+                }
+                if (reasoningSummary) {
+                    options.openai.reasoningSummary = reasoningSummary as
+                        | "none"
+                        | "brief"
+                        | "detailed"
+                }
+            }
+            break
+        }
+
+        case "anthropic": {
+            const thinkingBudget = parseIntSafe(
+                process.env.ANTHROPIC_THINKING_BUDGET_TOKENS,
+                "ANTHROPIC_THINKING_BUDGET_TOKENS",
+                1024,
+                64000,
+            )
+            const thinkingType =
+                process.env.ANTHROPIC_THINKING_TYPE || "enabled"
+
+            if (thinkingBudget) {
+                options.anthropic = {
+                    thinking: {
+                        type: thinkingType,
+                        budgetTokens: thinkingBudget,
+                    },
+                }
+            }
+            break
+        }
+
+        case "google": {
+            const reasoningEffort = process.env.GOOGLE_REASONING_EFFORT
+            const thinkingBudgetVal = parseIntSafe(
+                process.env.GOOGLE_THINKING_BUDGET,
+                "GOOGLE_THINKING_BUDGET",
+                1024,
+                100000,
+            )
+            const thinkingLevel = process.env.GOOGLE_THINKING_LEVEL
+
+            // Google Gemini 2.5/3 models think by default, but need includeThoughts: true
+            // to return the reasoning in the response
+            if (
+                modelId &&
+                (modelId.includes("gemini-2") ||
+                    modelId.includes("gemini-3") ||
+                    modelId.includes("gemini2") ||
+                    modelId.includes("gemini3"))
+            ) {
+                const thinkingConfig: Record<string, any> = {
+                    includeThoughts: true,
+                }
+
+                // Optionally configure thinking budget or level
+                if (
+                    thinkingBudgetVal &&
+                    (modelId.includes("2.5") || modelId.includes("2-5"))
+                ) {
+                    thinkingConfig.thinkingBudget = thinkingBudgetVal
+                } else if (
+                    thinkingLevel &&
+                    (modelId.includes("gemini-3") ||
+                        modelId.includes("gemini3"))
+                ) {
+                    thinkingConfig.thinkingLevel = thinkingLevel as
+                        | "low"
+                        | "high"
+                }
+
+                options.google = { thinkingConfig }
+            } else if (reasoningEffort) {
+                options.google = {
+                    reasoningEffort: reasoningEffort as
+                        | "low"
+                        | "medium"
+                        | "high",
+                }
+            }
+
+            // Keep existing Google options
+            const options_obj: Record<string, any> = {}
+            const candidateCount = parseIntSafe(
+                process.env.GOOGLE_CANDIDATE_COUNT,
+                "GOOGLE_CANDIDATE_COUNT",
+                1,
+                8,
+            )
+            if (candidateCount) {
+                options_obj.candidateCount = candidateCount
+            }
+            const topK = parseIntSafe(
+                process.env.GOOGLE_TOP_K,
+                "GOOGLE_TOP_K",
+                1,
+                100,
+            )
+            if (topK) {
+                options_obj.topK = topK
+            }
+            if (process.env.GOOGLE_TOP_P) {
+                const topP = Number.parseFloat(process.env.GOOGLE_TOP_P)
+                if (Number.isNaN(topP) || topP < 0 || topP > 1) {
+                    throw new Error(
+                        `GOOGLE_TOP_P must be a number between 0 and 1, got: ${process.env.GOOGLE_TOP_P}`,
+                    )
+                }
+                options_obj.topP = topP
+            }
+
+            if (Object.keys(options_obj).length > 0) {
+                options.google = { ...options.google, ...options_obj }
+            }
+            break
+        }
+
+        case "azure": {
+            const reasoningEffort = process.env.AZURE_REASONING_EFFORT
+            const reasoningSummary = process.env.AZURE_REASONING_SUMMARY
+
+            if (reasoningEffort || reasoningSummary) {
+                options.azure = {}
+                if (reasoningEffort) {
+                    options.azure.reasoningEffort = reasoningEffort as
+                        | "low"
+                        | "medium"
+                        | "high"
+                }
+                if (reasoningSummary) {
+                    options.azure.reasoningSummary = reasoningSummary as
+                        | "none"
+                        | "brief"
+                        | "detailed"
+                }
+            }
+            break
+        }
+
+        case "bedrock": {
+            const budgetTokens = parseIntSafe(
+                process.env.BEDROCK_REASONING_BUDGET_TOKENS,
+                "BEDROCK_REASONING_BUDGET_TOKENS",
+                1024,
+                64000,
+            )
+            const reasoningEffort = process.env.BEDROCK_REASONING_EFFORT
+
+            // Bedrock reasoning ONLY for Claude and Nova models
+            // Other models (MiniMax, etc.) don't support reasoningConfig
+            if (
+                modelId &&
+                (budgetTokens || reasoningEffort) &&
+                (modelId.includes("claude") ||
+                    modelId.includes("anthropic") ||
+                    modelId.includes("nova") ||
+                    modelId.includes("amazon"))
+            ) {
+                const reasoningConfig: Record<string, any> = { type: "enabled" }
+
+                // Claude models: use budgetTokens (1024-64000)
+                if (
+                    budgetTokens &&
+                    (modelId.includes("claude") ||
+                        modelId.includes("anthropic"))
+                ) {
+                    reasoningConfig.budgetTokens = budgetTokens
+                }
+                // Nova models: use maxReasoningEffort (low/medium/high)
+                else if (
+                    reasoningEffort &&
+                    (modelId.includes("nova") || modelId.includes("amazon"))
+                ) {
+                    reasoningConfig.maxReasoningEffort = reasoningEffort as
+                        | "low"
+                        | "medium"
+                        | "high"
+                }
+
+                options.bedrock = { reasoningConfig }
+            }
+            break
+        }
+
+        case "ollama": {
+            const enableThinking = process.env.OLLAMA_ENABLE_THINKING
+            // Ollama supports reasoning with think: true for models like qwen3
+            if (enableThinking === "true") {
+                options.ollama = { think: true }
+            }
+            break
+        }
+
+        case "deepseek":
+        case "openrouter":
+        case "siliconflow":
+        case "gateway": {
+            // These providers don't have reasoning configs in AI SDK yet
+            // Gateway passes through to underlying providers which handle their own configs
+            break
+        }
+
+        default:
+            break
+    }
+
+    return Object.keys(options).length > 0 ? options : undefined
+}
+
 // Map of provider to required environment variable
 const PROVIDER_ENV_VARS: Record<ProviderName, string | null> = {
     bedrock: null, // AWS SDK auto-uses IAM role on AWS, or env vars locally
@@ -47,6 +360,8 @@ const PROVIDER_ENV_VARS: Record<ProviderName, string | null> = {
     ollama: null, // No credentials needed for local Ollama
     openrouter: "OPENROUTER_API_KEY",
     deepseek: "DEEPSEEK_API_KEY",
+    siliconflow: "SILICONFLOW_API_KEY",
+    gateway: "AI_GATEWAY_API_KEY",
 }
 
 /**
@@ -62,7 +377,16 @@ function detectProvider(): ProviderName | null {
             continue
         }
         if (process.env[envVar]) {
-            configuredProviders.push(provider as ProviderName)
+            // Azure requires additional config (baseURL or resourceName)
+            if (provider === "azure") {
+                const hasBaseUrl = !!process.env.AZURE_BASE_URL
+                const hasResourceName = !!process.env.AZURE_RESOURCE_NAME
+                if (hasBaseUrl || hasResourceName) {
+                    configuredProviders.push(provider as ProviderName)
+                }
+            } else {
+                configuredProviders.push(provider as ProviderName)
+            }
         }
     }
 
@@ -84,13 +408,25 @@ function validateProviderCredentials(provider: ProviderName): void {
                 `Please set it in your .env.local file.`,
         )
     }
+
+    // Azure requires either AZURE_BASE_URL or AZURE_RESOURCE_NAME in addition to API key
+    if (provider === "azure") {
+        const hasBaseUrl = !!process.env.AZURE_BASE_URL
+        const hasResourceName = !!process.env.AZURE_RESOURCE_NAME
+        if (!hasBaseUrl && !hasResourceName) {
+            throw new Error(
+                `Azure requires either AZURE_BASE_URL or AZURE_RESOURCE_NAME to be set. ` +
+                    `Please set one in your .env.local file.`,
+            )
+        }
+    }
 }
 
 /**
  * Get the AI model based on environment variables
  *
  * Environment variables:
- * - AI_PROVIDER: The provider to use (bedrock, openai, anthropic, google, azure, ollama, openrouter, deepseek)
+ * - AI_PROVIDER: The provider to use (bedrock, openai, anthropic, google, azure, ollama, openrouter, deepseek, siliconflow)
  * - AI_MODEL: The model ID/name for the selected provider
  *
  * Provider-specific env vars:
@@ -104,19 +440,52 @@ function validateProviderCredentials(provider: ProviderName): void {
  * - OPENROUTER_API_KEY: OpenRouter API key
  * - DEEPSEEK_API_KEY: DeepSeek API key
  * - DEEPSEEK_BASE_URL: DeepSeek endpoint (optional)
+ * - SILICONFLOW_API_KEY: SiliconFlow API key
+ * - SILICONFLOW_BASE_URL: SiliconFlow endpoint (optional, defaults to https://api.siliconflow.com/v1)
  */
-export function getAIModel(): ModelConfig {
-    const modelId = process.env.AI_MODEL
+export function getAIModel(overrides?: ClientOverrides): ModelConfig {
+    // SECURITY: Prevent SSRF attacks (GHSA-9qf7-mprq-9qgm)
+    // If a custom baseUrl is provided, an API key MUST also be provided.
+    // This prevents attackers from redirecting server API keys to malicious endpoints.
+    if (overrides?.baseUrl && !overrides?.apiKey) {
+        throw new Error(
+            `API key is required when using a custom base URL. ` +
+                `Please provide your own API key in Settings.`,
+        )
+    }
+
+    // Check if client is providing their own provider override
+    const isClientOverride = !!(overrides?.provider && overrides?.apiKey)
+
+    // Use client override if provided, otherwise fall back to env vars
+    const modelId = overrides?.modelId || process.env.AI_MODEL
 
     if (!modelId) {
+        if (isClientOverride) {
+            throw new Error(
+                `Model ID is required when using custom AI provider. Please specify a model in Settings.`,
+            )
+        }
         throw new Error(
             `AI_MODEL environment variable is required. Example: AI_MODEL=claude-sonnet-4-5`,
         )
     }
 
-    // Determine provider: explicit config > auto-detect > error
+    // Determine provider: client override > explicit config > auto-detect > error
     let provider: ProviderName
-    if (process.env.AI_PROVIDER) {
+    if (overrides?.provider) {
+        // Validate client-provided provider
+        if (
+            !ALLOWED_CLIENT_PROVIDERS.includes(
+                overrides.provider as ProviderName,
+            )
+        ) {
+            throw new Error(
+                `Invalid provider: ${overrides.provider}. Allowed providers: ${ALLOWED_CLIENT_PROVIDERS.join(", ")}`,
+            )
+        }
+        provider = overrides.provider as ProviderName
+    } else if (process.env.AI_PROVIDER) {
         provider = process.env.AI_PROVIDER as ProviderName
     } else {
         const detected = detectProvider()
@@ -132,6 +501,7 @@ export function getAIModel(): ModelConfig {
             if (configured.length === 0) {
                 throw new Error(
                     `No AI provider configured. Please set one of the following API keys in your .env.local file:\n` +
+                        `- AI_GATEWAY_API_KEY for Vercel AI Gateway\n` +
                         `- DEEPSEEK_API_KEY for DeepSeek\n` +
                         `- OPENAI_API_KEY for OpenAI\n` +
                         `- ANTHROPIC_API_KEY for Anthropic\n` +
@@ -139,6 +509,7 @@ export function getAIModel(): ModelConfig {
                         `- AWS_ACCESS_KEY_ID for Bedrock\n` +
                         `- OPENROUTER_API_KEY for OpenRouter\n` +
                         `- AZURE_API_KEY for Azure\n` +
+                        `- SILICONFLOW_API_KEY for SiliconFlow\n` +
                         `Or set AI_PROVIDER=ollama for local Ollama.`,
                 )
             } else {
@@ -150,8 +521,10 @@ export function getAIModel(): ModelConfig {
         }
     }
 
-    // Validate provider credentials
-    validateProviderCredentials(provider)
+    // Only validate server credentials if client isn't providing their own API key
+    if (!isClientOverride) {
+        validateProviderCredentials(provider)
+    }
 
     console.log(`[AI Provider] Initializing ${provider} with model: ${modelId}`)
 
@@ -159,9 +532,12 @@ export function getAIModel(): ModelConfig {
     let providerOptions: any
     let headers: Record<string, string> | undefined
 
+    // Build provider-specific options from environment variables
+    const customProviderOptions = buildProviderOptions(provider, modelId)
+
     switch (provider) {
         case "bedrock": {
-            // Use credential provider chain for IAM role support (Amplify, Lambda, etc.)
+            // Use credential provider chain for IAM role support (Lambda, EC2, etc.)
             // Falls back to env vars (AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY) for local dev
             const bedrockProvider = createAmazonBedrock({
                 region: process.env.AWS_REGION || "us-west-2",
@@ -170,29 +546,43 @@ export function getAIModel(): ModelConfig {
             model = bedrockProvider(modelId)
             // Add Anthropic beta options if using Claude models via Bedrock
             if (modelId.includes("anthropic.claude")) {
-                providerOptions = BEDROCK_ANTHROPIC_BETA
+                // Deep merge to preserve both anthropicBeta and reasoningConfig
+                providerOptions = {
+                    bedrock: {
+                        ...BEDROCK_ANTHROPIC_BETA.bedrock,
+                        ...(customProviderOptions?.bedrock || {}),
+                    },
+                }
+            } else if (customProviderOptions) {
+                providerOptions = customProviderOptions
             }
             break
         }
 
-        case "openai":
-            if (process.env.OPENAI_BASE_URL) {
+        case "openai": {
+            const apiKey = overrides?.apiKey || process.env.OPENAI_API_KEY
+            const baseURL = overrides?.baseUrl || process.env.OPENAI_BASE_URL
+            if (baseURL || overrides?.apiKey) {
                 const customOpenAI = createOpenAI({
-                    apiKey: process.env.OPENAI_API_KEY,
-                    baseURL: process.env.OPENAI_BASE_URL,
+                    apiKey,
+                    ...(baseURL && { baseURL }),
                 })
                 model = customOpenAI.chat(modelId)
             } else {
                 model = openai(modelId)
             }
             break
+        }
 
         case "anthropic": {
+            const apiKey = overrides?.apiKey || process.env.ANTHROPIC_API_KEY
+            const baseURL =
+                overrides?.baseUrl ||
+                process.env.ANTHROPIC_BASE_URL ||
+                "https://api.anthropic.com/v1"
             const customProvider = createAnthropic({
-                apiKey: process.env.ANTHROPIC_API_KEY,
-                baseURL:
-                    process.env.ANTHROPIC_BASE_URL ||
-                    "https://api.anthropic.com/v1",
+                apiKey,
+                baseURL,
                 headers: ANTHROPIC_BETA_HEADERS,
             })
             model = customProvider(modelId)
@@ -201,29 +591,41 @@ export function getAIModel(): ModelConfig {
             break
         }
 
-        case "google":
-            if (process.env.GOOGLE_BASE_URL) {
+        case "google": {
+            const apiKey =
+                overrides?.apiKey || process.env.GOOGLE_GENERATIVE_AI_API_KEY
+            const baseURL = overrides?.baseUrl || process.env.GOOGLE_BASE_URL
+            if (baseURL || overrides?.apiKey) {
                 const customGoogle = createGoogleGenerativeAI({
-                    apiKey: process.env.GOOGLE_GENERATIVE_AI_API_KEY,
-                    baseURL: process.env.GOOGLE_BASE_URL,
+                    apiKey,
+                    ...(baseURL && { baseURL }),
                 })
                 model = customGoogle(modelId)
             } else {
                 model = google(modelId)
             }
             break
+        }
 
-        case "azure":
-            if (process.env.AZURE_BASE_URL) {
+        case "azure": {
+            const apiKey = overrides?.apiKey || process.env.AZURE_API_KEY
+            const baseURL = overrides?.baseUrl || process.env.AZURE_BASE_URL
+            const resourceName = process.env.AZURE_RESOURCE_NAME
+            // Azure requires either baseURL or resourceName to construct the endpoint
+            // resourceName constructs: https://{resourceName}.openai.azure.com/openai/v1{path}
+            if (baseURL || resourceName || overrides?.apiKey) {
                 const customAzure = createAzure({
-                    apiKey: process.env.AZURE_API_KEY,
-                    baseURL: process.env.AZURE_BASE_URL,
+                    apiKey,
+                    // baseURL takes precedence over resourceName per SDK behavior
+                    ...(baseURL && { baseURL }),
+                    ...(!baseURL && resourceName && { resourceName }),
                 })
                 model = customAzure(modelId)
             } else {
                 model = azure(modelId)
             }
             break
+        }
 
         case "ollama":
             if (process.env.OLLAMA_BASE_URL) {
@@ -237,33 +639,78 @@ export function getAIModel(): ModelConfig {
             break
 
         case "openrouter": {
+            const apiKey = overrides?.apiKey || process.env.OPENROUTER_API_KEY
+            const baseURL =
+                overrides?.baseUrl || process.env.OPENROUTER_BASE_URL
             const openrouter = createOpenRouter({
-                apiKey: process.env.OPENROUTER_API_KEY,
-                ...(process.env.OPENROUTER_BASE_URL && {
-                    baseURL: process.env.OPENROUTER_BASE_URL,
-                }),
+                apiKey,
+                ...(baseURL && { baseURL }),
             })
             model = openrouter(modelId)
             break
         }
 
-        case "deepseek":
-            if (process.env.DEEPSEEK_BASE_URL) {
+        case "deepseek": {
+            const apiKey = overrides?.apiKey || process.env.DEEPSEEK_API_KEY
+            const baseURL = overrides?.baseUrl || process.env.DEEPSEEK_BASE_URL
+            if (baseURL || overrides?.apiKey) {
                 const customDeepSeek = createDeepSeek({
-                    apiKey: process.env.DEEPSEEK_API_KEY,
-                    baseURL: process.env.DEEPSEEK_BASE_URL,
+                    apiKey,
+                    ...(baseURL && { baseURL }),
                 })
                 model = customDeepSeek(modelId)
             } else {
                 model = deepseek(modelId)
             }
             break
+        }
+
+        case "siliconflow": {
+            const apiKey = overrides?.apiKey || process.env.SILICONFLOW_API_KEY
+            const baseURL =
+                overrides?.baseUrl ||
+                process.env.SILICONFLOW_BASE_URL ||
+                "https://api.siliconflow.com/v1"
+            const siliconflowProvider = createOpenAI({
+                apiKey,
+                baseURL,
+            })
+            model = siliconflowProvider.chat(modelId)
+            break
+        }
+
+        case "gateway": {
+            // Vercel AI Gateway - unified access to multiple AI providers
+            // Model format: "provider/model" e.g., "openai/gpt-4o", "anthropic/claude-sonnet-4-5"
+            // See: https://vercel.com/ai-gateway
+            model = gateway(modelId)
+            break
+        }
 
         default:
             throw new Error(
-                `Unknown AI provider: ${provider}. Supported providers: bedrock, openai, anthropic, google, azure, ollama, openrouter, deepseek`,
+                `Unknown AI provider: ${provider}. Supported providers: bedrock, openai, anthropic, google, azure, ollama, openrouter, deepseek, siliconflow, gateway`,
             )
     }
 
+    // Apply provider-specific options for all providers except bedrock (which has special handling)
+    if (customProviderOptions && provider !== "bedrock" && !providerOptions) {
+        providerOptions = customProviderOptions
+    }
+
     return { model, providerOptions, headers, modelId }
 }
+
+/**
+ * Check if a model supports prompt caching.
+ * Currently only Claude models on Bedrock support prompt caching.
+ */
+export function supportsPromptCaching(modelId: string): boolean {
+    // Bedrock prompt caching is supported for Claude models
+    return (
+        modelId.includes("claude") ||
+        modelId.includes("anthropic") ||
+        modelId.startsWith("us.anthropic") ||
+        modelId.startsWith("eu.anthropic")
+    )
+}

lib/cached-responses.ts

@@ -9,12 +9,7 @@ export const CACHED_EXAMPLE_RESPONSES: CachedResponse[] = [
         promptText:
             "Give me a **animated connector** diagram of transformer's architecture",
         hasImage: false,
-        xml: `<root>
-  <mxCell id="0"/>
-  <mxCell id="1" parent="0"/>
-
-
-  <mxCell id="title" value="Transformer Architecture" style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=20;fontStyle=1;" vertex="1" parent="1">
+        xml: `<mxCell id="title" value="Transformer Architecture" style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=20;fontStyle=1;" vertex="1" parent="1">
     <mxGeometry x="300" y="20" width="250" height="30" as="geometry"/>
   </mxCell>
 
@@ -254,18 +249,12 @@ export const CACHED_EXAMPLE_RESPONSES: CachedResponse[] = [
 
   <mxCell id="output_label" value="Outputs&#xa;(shifted right)" style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=12;fontStyle=1;" vertex="1" parent="1">
     <mxGeometry x="660" y="530" width="100" height="30" as="geometry"/>
-  </mxCell>
-</root>`,
+  </mxCell>`,
     },
     {
         promptText: "Replicate this in aws style",
         hasImage: true,
-        xml: `<root>
-  <mxCell id="0"/>
-  <mxCell id="1" parent="0"/>
-
-
-  <mxCell id="2" value="AWS" style="sketch=0;outlineConnect=0;gradientColor=none;html=1;whiteSpace=wrap;fontSize=12;fontStyle=0;container=1;pointerEvents=0;collapsible=0;recursiveResize=0;shape=mxgraph.aws4.group;grIcon=mxgraph.aws4.group_aws_cloud;strokeColor=#232F3E;fillColor=none;verticalAlign=top;align=left;spacingLeft=30;fontColor=#232F3E;dashed=0;rounded=1;arcSize=5;" vertex="1" parent="1">
+        xml: `<mxCell id="2" value="AWS" style="sketch=0;outlineConnect=0;gradientColor=none;html=1;whiteSpace=wrap;fontSize=12;fontStyle=0;container=1;pointerEvents=0;collapsible=0;recursiveResize=0;shape=mxgraph.aws4.group;grIcon=mxgraph.aws4.group_aws_cloud;strokeColor=#232F3E;fillColor=none;verticalAlign=top;align=left;spacingLeft=30;fontColor=#232F3E;dashed=0;rounded=1;arcSize=5;" vertex="1" parent="1">
     <mxGeometry x="340" y="40" width="880" height="520" as="geometry"/>
   </mxCell>
 
@@ -324,18 +313,12 @@ export const CACHED_EXAMPLE_RESPONSES: CachedResponse[] = [
       <mxPoint x="700" y="350" as="sourcePoint"/>
       <mxPoint x="750" y="300" as="targetPoint"/>
     </mxGeometry>
-  </mxCell>
-</root>`,
+  </mxCell>`,
     },
     {
         promptText: "Replicate this flowchart.",
         hasImage: true,
-        xml: `<root>
-  <mxCell id="0"/>
-  <mxCell id="1" parent="0"/>
-
-
-  <mxCell id="2" value="Lamp doesn't work" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#ffcccc;strokeColor=#000000;strokeWidth=2;fontSize=18;fontStyle=0;" vertex="1" parent="1">
+        xml: `<mxCell id="2" value="Lamp doesn't work" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#ffcccc;strokeColor=#000000;strokeWidth=2;fontSize=18;fontStyle=0;" vertex="1" parent="1">
     <mxGeometry x="140" y="40" width="180" height="60" as="geometry"/>
   </mxCell>
 
@@ -391,18 +374,368 @@ export const CACHED_EXAMPLE_RESPONSES: CachedResponse[] = [
 
   <mxCell id="12" value="Repair lamp" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#99ff99;strokeColor=#000000;strokeWidth=2;fontSize=18;fontStyle=0;" vertex="1" parent="1">
     <mxGeometry x="130" y="650" width="200" height="60" as="geometry"/>
-  </mxCell>
-</root>`,
+  </mxCell>`,
+    },
+    {
+        promptText: "Summarize this paper as a diagram",
+        hasImage: true,
+        xml: `<mxCell id="title_bg" parent="1"
+                    style="rounded=1;whiteSpace=wrap;html=1;fillColor=#1a237e;strokeColor=none;arcSize=8;"
+                    value="" vertex="1">
+                    <mxGeometry height="80" width="720" x="40" y="20" as="geometry" />
+                </mxCell>
+                <mxCell id="title" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=22;fontStyle=1;fontColor=#FFFFFF;"
+                    value="Chain-of-Thought Prompting&lt;br&gt;&lt;font style=&quot;font-size: 14px;&quot;&gt;Elicits Reasoning in Large Language Models&lt;/font&gt;"
+                    vertex="1">
+                    <mxGeometry height="70" width="720" x="40" y="25" as="geometry" />
+                </mxCell>
+                <mxCell id="authors" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=11;fontColor=#666666;"
+                    value="Wei et al. (Google Research, Brain Team) | NeurIPS 2022" vertex="1">
+                    <mxGeometry height="20" width="720" x="40" y="100" as="geometry" />
+                </mxCell>
+                <mxCell id="core_header" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=left;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=16;fontStyle=1;fontColor=#1a237e;"
+                    value="💡 Core Idea" vertex="1">
+                    <mxGeometry height="30" width="150" x="40" y="125" as="geometry" />
+                </mxCell>
+                <mxCell id="core_box" parent="1"
+                    style="rounded=1;whiteSpace=wrap;html=1;fillColor=#E3F2FD;strokeColor=#1565C0;align=left;spacingLeft=10;spacingRight=10;fontSize=11;"
+                    value="&lt;b&gt;Chain of Thought&lt;/b&gt; = A series of intermediate reasoning steps that lead to the final answer&lt;br&gt;&lt;br&gt;Simply provide a few CoT demonstrations as exemplars in few-shot prompting"
+                    vertex="1">
+                    <mxGeometry height="75" width="340" x="40" y="155" as="geometry" />
+                </mxCell>
+                <mxCell id="compare_header" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=left;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=16;fontStyle=1;fontColor=#1a237e;"
+                    value="⚖️ Standard vs Chain-of-Thought Prompting" vertex="1">
+                    <mxGeometry height="30" width="350" x="40" y="240" as="geometry" />
+                </mxCell>
+                <mxCell id="std_box" parent="1"
+                    style="rounded=1;whiteSpace=wrap;html=1;fillColor=#FFEBEE;strokeColor=#C62828;arcSize=8;"
+                    value="" vertex="1">
+                    <mxGeometry height="160" width="170" x="40" y="275" as="geometry" />
+                </mxCell>
+                <mxCell id="std_title" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=12;fontStyle=1;fontColor=#C62828;"
+                    value="Standard Prompting" vertex="1">
+                    <mxGeometry height="25" width="170" x="40" y="280" as="geometry" />
+                </mxCell>
+                <mxCell id="std_q" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=left;verticalAlign=top;whiteSpace=wrap;rounded=0;fontSize=9;spacingLeft=5;spacingRight=5;"
+                    value="Q: Roger has 5 tennis balls. He buys 2 more cans. Each can has 3 balls. How many now?"
+                    vertex="1">
+                    <mxGeometry height="55" width="160" x="45" y="305" as="geometry" />
+                </mxCell>
+                <mxCell id="std_a" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=#FFCDD2;align=left;verticalAlign=middle;whiteSpace=wrap;rounded=1;fontSize=10;fontStyle=1;spacingLeft=5;"
+                    value="A: The answer is 11." vertex="1">
+                    <mxGeometry height="25" width="150" x="50" y="365" as="geometry" />
+                </mxCell>
+                <mxCell id="std_result" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=11;fontStyle=1;fontColor=#C62828;"
+                    value="❌ Often Wrong" vertex="1">
+                    <mxGeometry height="30" width="170" x="40" y="400" as="geometry" />
+                </mxCell>
+                <mxCell id="cot_box" parent="1"
+                    style="rounded=1;whiteSpace=wrap;html=1;fillColor=#E8F5E9;strokeColor=#2E7D32;arcSize=8;"
+                    value="" vertex="1">
+                    <mxGeometry height="160" width="170" x="220" y="275" as="geometry" />
+                </mxCell>
+                <mxCell id="cot_title" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=12;fontStyle=1;fontColor=#2E7D32;"
+                    value="Chain-of-Thought" vertex="1">
+                    <mxGeometry height="25" width="170" x="220" y="280" as="geometry" />
+                </mxCell>
+                <mxCell id="cot_q" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=left;verticalAlign=top;whiteSpace=wrap;rounded=0;fontSize=9;spacingLeft=5;spacingRight=5;"
+                    value="Q: Roger has 5 tennis balls. He buys 2 more cans. Each can has 3 balls. How many now?"
+                    vertex="1">
+                    <mxGeometry height="55" width="160" x="225" y="305" as="geometry" />
+                </mxCell>
+                <mxCell id="cot_a" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=#C8E6C9;align=left;verticalAlign=middle;whiteSpace=wrap;rounded=1;fontSize=9;fontStyle=1;spacingLeft=5;"
+                    value="A: 2 cans × 3 = 6 balls.&lt;br&gt;5 + 6 = 11. Answer: 11" vertex="1">
+                    <mxGeometry height="35" width="150" x="230" y="360" as="geometry" />
+                </mxCell>
+                <mxCell id="cot_result" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=11;fontStyle=1;fontColor=#2E7D32;"
+                    value="✓ Correct!" vertex="1">
+                    <mxGeometry height="30" width="170" x="220" y="400" as="geometry" />
+                </mxCell>
+                <mxCell id="vs_arrow" edge="1" parent="1"
+                    style="shape=flexArrow;endArrow=classic;startArrow=classic;html=1;fillColor=#FFC107;strokeColor=none;width=8;endSize=4;startSize=4;"
+                    value="">
+                    <mxGeometry relative="1" width="100" as="geometry">
+                        <mxPoint x="195" y="355" as="sourcePoint" />
+                        <mxPoint x="235" y="355" as="targetPoint" />
+                    </mxGeometry>
+                </mxCell>
+                <mxCell id="props_header" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=left;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=16;fontStyle=1;fontColor=#1a237e;"
+                    value="🔑 Key Properties" vertex="1">
+                    <mxGeometry height="30" width="150" x="400" y="125" as="geometry" />
+                </mxCell>
+                <mxCell id="prop1" parent="1"
+                    style="rounded=1;whiteSpace=wrap;html=1;fillColor=#FFF3E0;strokeColor=#EF6C00;fontSize=10;align=left;spacingLeft=8;"
+                    value="1️⃣ Decomposes multi-step problems" vertex="1">
+                    <mxGeometry height="32" width="180" x="400" y="155" as="geometry" />
+                </mxCell>
+                <mxCell id="prop2" parent="1"
+                    style="rounded=1;whiteSpace=wrap;html=1;fillColor=#FFF3E0;strokeColor=#EF6C00;fontSize=10;align=left;spacingLeft=8;"
+                    value="2️⃣ Interpretable reasoning window" vertex="1">
+                    <mxGeometry height="32" width="180" x="400" y="192" as="geometry" />
+                </mxCell>
+                <mxCell id="prop3" parent="1"
+                    style="rounded=1;whiteSpace=wrap;html=1;fillColor=#FFF3E0;strokeColor=#EF6C00;fontSize=10;align=left;spacingLeft=8;"
+                    value="3️⃣ Applicable to any language task" vertex="1">
+                    <mxGeometry height="32" width="180" x="400" y="229" as="geometry" />
+                </mxCell>
+                <mxCell id="prop4" parent="1"
+                    style="rounded=1;whiteSpace=wrap;html=1;fillColor=#FFF3E0;strokeColor=#EF6C00;fontSize=10;align=left;spacingLeft=8;"
+                    value="4️⃣ No finetuning required" vertex="1">
+                    <mxGeometry height="32" width="180" x="400" y="266" as="geometry" />
+                </mxCell>
+                <mxCell id="emergent_header" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=left;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=16;fontStyle=1;fontColor=#1a237e;"
+                    value="📈 Emergent Ability" vertex="1">
+                    <mxGeometry height="30" width="180" x="400" y="310" as="geometry" />
+                </mxCell>
+                <mxCell id="emergent_box" parent="1"
+                    style="rounded=1;whiteSpace=wrap;html=1;fillColor=#F3E5F5;strokeColor=#7B1FA2;arcSize=8;"
+                    value="" vertex="1">
+                    <mxGeometry height="95" width="180" x="400" y="340" as="geometry" />
+                </mxCell>
+                <mxCell id="emergent_text" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=11;"
+                    value="CoT only works with&lt;br&gt;&lt;b&gt;~100B+ parameters&lt;/b&gt;&lt;br&gt;&lt;br&gt;Small models produce&lt;br&gt;fluent but illogical chains"
+                    vertex="1">
+                    <mxGeometry height="85" width="180" x="400" y="345" as="geometry" />
+                </mxCell>
+                <mxCell id="results_header" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=left;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=16;fontStyle=1;fontColor=#1a237e;"
+                    value="📊 Key Results" vertex="1">
+                    <mxGeometry height="30" width="150" x="600" y="125" as="geometry" />
+                </mxCell>
+                <mxCell id="gsm_box" parent="1"
+                    style="rounded=1;whiteSpace=wrap;html=1;fillColor=#E8F5E9;strokeColor=#2E7D32;arcSize=8;"
+                    value="" vertex="1">
+                    <mxGeometry height="100" width="160" x="600" y="155" as="geometry" />
+                </mxCell>
+                <mxCell id="gsm_title" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=12;fontStyle=1;fontColor=#2E7D32;"
+                    value="GSM8K (Math)" vertex="1">
+                    <mxGeometry height="20" width="160" x="600" y="160" as="geometry" />
+                </mxCell>
+                <mxCell id="gsm_bar1" parent="1"
+                    style="rounded=0;whiteSpace=wrap;html=1;fillColor=#FFCDD2;strokeColor=none;"
+                    value="" vertex="1">
+                    <mxGeometry height="30" width="40" x="615" y="185" as="geometry" />
+                </mxCell>
+                <mxCell id="gsm_bar2" parent="1"
+                    style="rounded=0;whiteSpace=wrap;html=1;fillColor=#4CAF50;strokeColor=none;"
+                    value="" vertex="1">
+                    <mxGeometry height="30" width="80" x="665" y="185" as="geometry" />
+                </mxCell>
+                <mxCell id="gsm_label1" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=10;fontStyle=1;"
+                    value="18%" vertex="1">
+                    <mxGeometry height="15" width="40" x="615" y="215" as="geometry" />
+                </mxCell>
+                <mxCell id="gsm_label2" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=10;fontStyle=1;fontColor=#2E7D32;"
+                    value="57%" vertex="1">
+                    <mxGeometry height="15" width="80" x="665" y="215" as="geometry" />
+                </mxCell>
+                <mxCell id="gsm_legend" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=9;fontColor=#666666;"
+                    value="Standard → CoT (PaLM 540B)" vertex="1">
+                    <mxGeometry height="20" width="160" x="600" y="232" as="geometry" />
+                </mxCell>
+                <mxCell id="bench_header" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=left;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=16;fontStyle=1;fontColor=#1a237e;"
+                    value="🧪 Benchmarks Tested" vertex="1">
+                    <mxGeometry height="30" width="180" x="600" y="265" as="geometry" />
+                </mxCell>
+                <mxCell id="bench_arith" parent="1"
+                    style="rounded=1;whiteSpace=wrap;html=1;fillColor=#E3F2FD;strokeColor=#1565C0;fontSize=10;align=center;"
+                    value="🔢 Arithmetic&lt;br&gt;&lt;font style=&quot;font-size: 9px;&quot;&gt;GSM8K, SVAMP, ASDiv, AQuA, MAWPS&lt;/font&gt;"
+                    vertex="1">
+                    <mxGeometry height="45" width="160" x="600" y="295" as="geometry" />
+                </mxCell>
+                <mxCell id="bench_common" parent="1"
+                    style="rounded=1;whiteSpace=wrap;html=1;fillColor=#E3F2FD;strokeColor=#1565C0;fontSize=10;align=center;"
+                    value="🧠 Commonsense&lt;br&gt;&lt;font style=&quot;font-size: 9px;&quot;&gt;CSQA, StrategyQA, Date, Sports, SayCan&lt;/font&gt;"
+                    vertex="1">
+                    <mxGeometry height="45" width="160" x="600" y="345" as="geometry" />
+                </mxCell>
+                <mxCell id="bench_symbol" parent="1"
+                    style="rounded=1;whiteSpace=wrap;html=1;fillColor=#E3F2FD;strokeColor=#1565C0;fontSize=10;align=center;"
+                    value="🔣 Symbolic&lt;br&gt;&lt;font style=&quot;font-size: 9px;&quot;&gt;Last Letter Concat, Coin Flip&lt;/font&gt;"
+                    vertex="1">
+                    <mxGeometry height="40" width="160" x="600" y="395" as="geometry" />
+                </mxCell>
+                <mxCell id="task_header" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=left;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=16;fontStyle=1;fontColor=#1a237e;"
+                    value="🎯 Task Types &amp; Results" vertex="1">
+                    <mxGeometry height="30" width="200" x="40" y="445" as="geometry" />
+                </mxCell>
+                <mxCell id="task_arith" parent="1"
+                    style="ellipse;whiteSpace=wrap;html=1;fillColor=#BBDEFB;strokeColor=#1565C0;fontSize=11;fontStyle=1;"
+                    value="Arithmetic&lt;br&gt;Reasoning" vertex="1">
+                    <mxGeometry height="60" width="90" x="40" y="480" as="geometry" />
+                </mxCell>
+                <mxCell id="task_arith_res" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=top;whiteSpace=wrap;rounded=0;fontSize=9;fontColor=#1565C0;"
+                    value="SOTA on GSM8K&lt;br&gt;(57% vs 55% prior)" vertex="1">
+                    <mxGeometry height="30" width="110" x="30" y="540" as="geometry" />
+                </mxCell>
+                <mxCell id="task_common" parent="1"
+                    style="ellipse;whiteSpace=wrap;html=1;fillColor=#C8E6C9;strokeColor=#2E7D32;fontSize=11;fontStyle=1;"
+                    value="Commonsense&lt;br&gt;Reasoning" vertex="1">
+                    <mxGeometry height="60" width="90" x="160" y="480" as="geometry" />
+                </mxCell>
+                <mxCell id="task_common_res" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=top;whiteSpace=wrap;rounded=0;fontSize=9;fontColor=#2E7D32;"
+                    value="SOTA StrategyQA&lt;br&gt;(75.6% vs 69.4%)" vertex="1">
+                    <mxGeometry height="30" width="110" x="150" y="540" as="geometry" />
+                </mxCell>
+                <mxCell id="task_symbol" parent="1"
+                    style="ellipse;whiteSpace=wrap;html=1;fillColor=#FFE0B2;strokeColor=#EF6C00;fontSize=11;fontStyle=1;"
+                    value="Symbolic&lt;br&gt;Reasoning" vertex="1">
+                    <mxGeometry height="60" width="90" x="280" y="480" as="geometry" />
+                </mxCell>
+                <mxCell id="task_symbol_res" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=top;whiteSpace=wrap;rounded=0;fontSize=9;fontColor=#EF6C00;"
+                    value="OOD Generalization&lt;br&gt;to longer sequences" vertex="1">
+                    <mxGeometry height="30" width="110" x="270" y="540" as="geometry" />
+                </mxCell>
+                <mxCell id="task_arrow1" edge="1" parent="1"
+                    style="endArrow=classic;html=1;strokeColor=#9E9E9E;strokeWidth=2;" value="">
+                    <mxGeometry height="50" relative="1" width="50" as="geometry">
+                        <mxPoint x="130" y="510" as="sourcePoint" />
+                        <mxPoint x="160" y="510" as="targetPoint" />
+                    </mxGeometry>
+                </mxCell>
+                <mxCell id="task_arrow2" edge="1" parent="1"
+                    style="endArrow=classic;html=1;strokeColor=#9E9E9E;strokeWidth=2;" value="">
+                    <mxGeometry height="50" relative="1" width="50" as="geometry">
+                        <mxPoint x="250" y="510" as="sourcePoint" />
+                        <mxPoint x="280" y="510" as="targetPoint" />
+                    </mxGeometry>
+                </mxCell>
+                <mxCell id="models_header" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=left;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=16;fontStyle=1;fontColor=#1a237e;"
+                    value="🤖 Models Tested" vertex="1">
+                    <mxGeometry height="30" width="150" x="400" y="445" as="geometry" />
+                </mxCell>
+                <mxCell id="models_box" parent="1"
+                    style="rounded=1;whiteSpace=wrap;html=1;fillColor=#ECEFF1;strokeColor=#607D8B;arcSize=8;"
+                    value="" vertex="1">
+                    <mxGeometry height="95" width="180" x="400" y="475" as="geometry" />
+                </mxCell>
+                <mxCell id="model1" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=left;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=11;spacingLeft=10;"
+                    value="• GPT-3 (175B)" vertex="1">
+                    <mxGeometry height="20" width="90" x="400" y="480" as="geometry" />
+                </mxCell>
+                <mxCell id="model2" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=left;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=11;spacingLeft=10;"
+                    value="• LaMDA (137B)" vertex="1">
+                    <mxGeometry height="20" width="90" x="400" y="500" as="geometry" />
+                </mxCell>
+                <mxCell id="model3" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=left;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=11;spacingLeft=10;"
+                    value="• PaLM (540B)" vertex="1">
+                    <mxGeometry height="20" width="90" x="400" y="520" as="geometry" />
+                </mxCell>
+                <mxCell id="model4" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=left;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=11;spacingLeft=10;"
+                    value="• Codex" vertex="1">
+                    <mxGeometry height="20" width="80" x="490" y="480" as="geometry" />
+                </mxCell>
+                <mxCell id="model5" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=left;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=11;spacingLeft=10;"
+                    value="• UL2 (20B)" vertex="1">
+                    <mxGeometry height="20" width="80" x="490" y="500" as="geometry" />
+                </mxCell>
+                <mxCell id="model_note" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=10;fontStyle=2;fontColor=#607D8B;"
+                    value="No finetuning - prompting only!" vertex="1">
+                    <mxGeometry height="20" width="180" x="400" y="545" as="geometry" />
+                </mxCell>
+                <mxCell id="takeaway_header" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=left;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=16;fontStyle=1;fontColor=#1a237e;"
+                    value="✨ Key Takeaways" vertex="1">
+                    <mxGeometry height="30" width="160" x="600" y="445" as="geometry" />
+                </mxCell>
+                <mxCell id="takeaway_box" parent="1"
+                    style="rounded=1;whiteSpace=wrap;html=1;fillColor=#FFF8E1;strokeColor=#FFA000;arcSize=8;"
+                    value="" vertex="1">
+                    <mxGeometry height="95" width="160" x="600" y="475" as="geometry" />
+                </mxCell>
+                <mxCell id="take1" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=left;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=10;spacingLeft=5;"
+                    value="✓ Simple yet powerful" vertex="1">
+                    <mxGeometry height="18" width="150" x="605" y="480" as="geometry" />
+                </mxCell>
+                <mxCell id="take2" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=left;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=10;spacingLeft=5;"
+                    value="✓ Emergent at scale" vertex="1">
+                    <mxGeometry height="18" width="150" x="605" y="498" as="geometry" />
+                </mxCell>
+                <mxCell id="take3" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=left;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=10;spacingLeft=5;"
+                    value="✓ Broadly applicable" vertex="1">
+                    <mxGeometry height="18" width="150" x="605" y="516" as="geometry" />
+                </mxCell>
+                <mxCell id="take4" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=left;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=10;spacingLeft=5;"
+                    value="✓ No training needed" vertex="1">
+                    <mxGeometry height="18" width="150" x="605" y="534" as="geometry" />
+                </mxCell>
+                <mxCell id="take5" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=left;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=10;spacingLeft=5;"
+                    value="✓ State-of-the-art results" vertex="1">
+                    <mxGeometry height="18" width="150" x="605" y="552" as="geometry" />
+                </mxCell>
+                <mxCell id="format_header" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=left;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=14;fontStyle=1;fontColor=#1a237e;"
+                    value="📝 Prompt Format" vertex="1">
+                    <mxGeometry height="25" width="150" x="40" y="575" as="geometry" />
+                </mxCell>
+                <mxCell id="format_box" parent="1"
+                    style="rounded=1;whiteSpace=wrap;html=1;fillColor=#E1BEE7;strokeColor=#7B1FA2;fontSize=12;fontStyle=1;"
+                    value="〈 Input, Chain of Thought, Output 〉" vertex="1">
+                    <mxGeometry height="35" width="250" x="40" y="600" as="geometry" />
+                </mxCell>
+                <mxCell id="limit_header" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=left;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=14;fontStyle=1;fontColor=#1a237e;"
+                    value="⚠️ Limitations" vertex="1">
+                    <mxGeometry height="25" width="120" x="310" y="575" as="geometry" />
+                </mxCell>
+                <mxCell id="limit_box" parent="1"
+                    style="rounded=1;whiteSpace=wrap;html=1;fillColor=#FFEBEE;strokeColor=#C62828;fontSize=10;align=left;spacingLeft=8;"
+                    value="• Requires large models (~100B+)&lt;br&gt;• No guarantee of correct reasoning&lt;br&gt;• Costly to serve in production"
+                    vertex="1">
+                    <mxGeometry height="55" width="200" x="310" y="600" as="geometry" />
+                </mxCell>
+                <mxCell id="impact_header" parent="1"
+                    style="text;html=1;strokeColor=none;fillColor=none;align=left;verticalAlign=middle;whiteSpace=wrap;rounded=0;fontSize=14;fontStyle=1;fontColor=#1a237e;"
+                    value="🚀 Impact" vertex="1">
+                    <mxGeometry height="25" width="100" x="530" y="575" as="geometry" />
+                </mxCell>
+                <mxCell id="impact_box" parent="1"
+                    style="rounded=1;whiteSpace=wrap;html=1;fillColor=#E8F5E9;strokeColor=#2E7D32;fontSize=10;align=left;spacingLeft=8;spacingRight=8;"
+                    value="Foundational technique for modern LLM reasoning - inspired many follow-up works including Self-Consistency, Tree-of-Thought, etc."
+                    vertex="1">
+                    <mxGeometry height="55" width="230" x="530" y="600" as="geometry" />
+                </mxCell>`,
     },
     {
         promptText: "Draw a cat for me",
         hasImage: false,
-        xml: `<root>
-  <mxCell id="0"/>
-  <mxCell id="1" parent="0"/>
-
-
-  <mxCell id="2" value="" style="ellipse;whiteSpace=wrap;html=1;aspect=fixed;fillColor=#FFE6CC;strokeColor=#000000;strokeWidth=2;" vertex="1" parent="1">
+        xml: `<mxCell id="2" value="" style="ellipse;whiteSpace=wrap;html=1;aspect=fixed;fillColor=#FFE6CC;strokeColor=#000000;strokeWidth=2;" vertex="1" parent="1">
     <mxGeometry x="300" y="150" width="120" height="120" as="geometry"/>
   </mxCell>
 
@@ -542,9 +875,7 @@ export const CACHED_EXAMPLE_RESPONSES: CachedResponse[] = [
         <mxPoint x="235" y="290"/>
       </Array>
     </mxGeometry>
-  </mxCell>
-
-</root>`,
+  </mxCell>`,
     },
 ]
 

lib/langfuse.ts

@@ -84,9 +84,7 @@ export function getTelemetryConfig(params: {
 
     return {
         isEnabled: true,
-        // Disable automatic input recording to avoid uploading large base64 images to Langfuse media
-        // User text input is recorded manually via setTraceInput
-        recordInputs: false,
+        recordInputs: true,
         recordOutputs: true,
         metadata: {
             sessionId: params.sessionId,

lib/pdf-utils.ts

@@ -0,0 +1,75 @@
+import { extractText, getDocumentProxy } from "unpdf"
+
+// Maximum characters allowed for extracted text (configurable via env)
+const DEFAULT_MAX_EXTRACTED_CHARS = 150000 // 150k chars
+export const MAX_EXTRACTED_CHARS =
+    Number(process.env.NEXT_PUBLIC_MAX_EXTRACTED_CHARS) ||
+    DEFAULT_MAX_EXTRACTED_CHARS
+
+// Text file extensions we support
+const TEXT_EXTENSIONS = [
+    ".txt",
+    ".md",
+    ".markdown",
+    ".json",
+    ".csv",
+    ".xml",
+    ".html",
+    ".css",
+    ".js",
+    ".ts",
+    ".jsx",
+    ".tsx",
+    ".py",
+    ".java",
+    ".c",
+    ".cpp",
+    ".h",
+    ".go",
+    ".rs",
+    ".yaml",
+    ".yml",
+    ".toml",
+    ".ini",
+    ".log",
+    ".sh",
+    ".bash",
+    ".zsh",
+]
+
+/**
+ * Extract text content from a PDF file
+ * Uses unpdf library for client-side extraction
+ */
+export async function extractPdfText(file: File): Promise<string> {
+    const buffer = await file.arrayBuffer()
+    const pdf = await getDocumentProxy(new Uint8Array(buffer))
+    const { text } = await extractText(pdf, { mergePages: true })
+    return text as string
+}
+
+/**
+ * Check if a file is a PDF
+ */
+export function isPdfFile(file: File): boolean {
+    return file.type === "application/pdf" || file.name.endsWith(".pdf")
+}
+
+/**
+ * Check if a file is a text file
+ */
+export function isTextFile(file: File): boolean {
+    const name = file.name.toLowerCase()
+    return (
+        file.type.startsWith("text/") ||
+        file.type === "application/json" ||
+        TEXT_EXTENSIONS.some((ext) => name.endsWith(ext))
+    )
+}
+
+/**
+ * Extract text content from a text file
+ */
+export async function extractTextFileContent(file: File): Promise<string> {
+    return await file.text()
+}

lib/storage.ts

@@ -0,0 +1,27 @@
+// Centralized localStorage keys
+// Consolidates all storage keys from chat-panel.tsx and settings-dialog.tsx
+
+export const STORAGE_KEYS = {
+    // Chat data
+    messages: "next-ai-draw-io-messages",
+    xmlSnapshots: "next-ai-draw-io-xml-snapshots",
+    diagramXml: "next-ai-draw-io-diagram-xml",
+    sessionId: "next-ai-draw-io-session-id",
+
+    // Quota tracking
+    requestCount: "next-ai-draw-io-request-count",
+    requestDate: "next-ai-draw-io-request-date",
+    tokenCount: "next-ai-draw-io-token-count",
+    tokenDate: "next-ai-draw-io-token-date",
+    tpmCount: "next-ai-draw-io-tpm-count",
+    tpmMinute: "next-ai-draw-io-tpm-minute",
+
+    // Settings
+    accessCode: "next-ai-draw-io-access-code",
+    closeProtection: "next-ai-draw-io-close-protection",
+    accessCodeRequired: "next-ai-draw-io-access-code-required",
+    aiProvider: "next-ai-draw-io-ai-provider",
+    aiBaseUrl: "next-ai-draw-io-ai-base-url",
+    aiApiKey: "next-ai-draw-io-ai-api-key",
+    aiModel: "next-ai-draw-io-ai-model",
+} as const

lib/system-prompts.ts

@@ -1,13 +1,19 @@
 /**
  * System prompts for different AI models
  * Extended prompt is used for models with higher cache token minimums (Opus 4.5, Haiku 4.5)
+ *
+ * Token counting utilities are in a separate file (token-counter.ts) to avoid
+ * WebAssembly issues with Next.js server-side rendering.
  */
 
-// Default system prompt (~1400 tokens) - works with all models
+// Default system prompt (~1900 tokens) - works with all models
 export const DEFAULT_SYSTEM_PROMPT = `
 You are an expert diagram creation assistant specializing in draw.io XML generation.
 Your primary function is chat with user and crafting clear, well-organized visual diagrams through precise XML specifications.
-You can see the image that user uploaded.
+You can see images that users upload, and you can read the text content extracted from PDF documents they upload.
+
+When you are asked to create a diagram, briefly describe your plan about the layout and structure to avoid object overlapping or edge cross the objects. (2-3 sentences max), then use display_diagram tool to generate the XML.
+After generating or editing a diagram, you don't need to say anything. The user can see the diagram - no need to describe it.
 
 ## App Context
 You are an AI agent (powered by {{MODEL_NAME}}) inside a web app. The interface has:
@@ -19,7 +25,7 @@ You can read and modify diagrams by generating draw.io XML code through tool cal
 ## App Features
 1. **Diagram History** (clock icon, bottom-left of chat input): The app automatically saves a snapshot before each AI edit. Users can view the history panel and restore any previous version. Feel free to make changes - nothing is permanently lost.
 2. **Theme Toggle** (palette icon, bottom-left of chat input): Users can switch between minimal UI and sketch-style UI for the draw.io editor.
-3. **Image Upload** (paperclip icon, bottom-left of chat input): Users can upload images for you to analyze and replicate as diagrams.
+3. **Image/PDF Upload** (paperclip icon, bottom-left of chat input): Users can upload images or PDF documents for you to analyze and generate diagrams from.
 4. **Export** (via draw.io toolbar): Users can save diagrams as .drawio, .svg, or .png files.
 5. **Clear Chat** (trash icon, bottom-right of chat input): Clears the conversation and resets the diagram.
 
@@ -36,11 +42,18 @@ description: Edit specific parts of the EXISTING diagram. Use this when making s
 parameters: {
   edits: Array<{search: string, replace: string}>
 }
+---Tool3---
+tool name: append_diagram
+description: Continue generating diagram XML when display_diagram was truncated due to output length limits. Only use this after display_diagram truncation.
+parameters: {
+  xml: string  // Continuation fragment (NO wrapper tags like <mxGraphModel> or <root>)
+}
 ---End of tools---
 
 IMPORTANT: Choose the right tool:
 - Use display_diagram for: Creating new diagrams, major restructuring, or when the current diagram XML is empty
 - Use edit_diagram for: Small modifications, adding/removing elements, changing text/colors, repositioning items
+- Use append_diagram for: ONLY when display_diagram was truncated due to output length - continue generating from where you stopped
 
 Core capabilities:
 - Generate valid, well-formed XML strings for draw.io diagrams
@@ -51,6 +64,8 @@ Core capabilities:
 - Optimize element positioning to prevent overlapping and maintain readability
 - Structure complex systems into clear, organized visual components
 
+
+
 Layout constraints:
 - CRITICAL: Keep all diagram elements within a single page viewport to avoid page breaks
 - Position all elements with x coordinates between 0-800 and y coordinates between 0-600
@@ -73,33 +88,33 @@ Note that:
 - NEVER include XML comments (<!-- ... -->) in your generated XML. Draw.io strips comments, which breaks edit_diagram patterns.
 
 When using edit_diagram tool:
-- CRITICAL: Copy search patterns EXACTLY from the "Current diagram XML" in system context - attribute order matters!
-- Always include the element's id attribute for unique targeting: {"search": "<mxCell id=\\"5\\"", ...}
-- Include complete elements (mxCell + mxGeometry) for reliable matching
-- Preserve exact whitespace, indentation, and line breaks
-- BAD: {"search": "value=\\"Label\\"", ...} - too vague, matches multiple elements
-- GOOD: {"search": "<mxCell id=\\"3\\" value=\\"Old\\" style=\\"...\\">", "replace": "<mxCell id=\\"3\\" value=\\"New\\" style=\\"...\\">"}
-- For multiple changes, use separate edits in array
-- RETRY POLICY: If pattern not found, retry up to 3 times with adjusted patterns. After 3 failures, use display_diagram instead.
+- Use operations: update (modify cell by id), add (new cell), delete (remove cell by id)
+- For update/add: provide cell_id and complete new_xml (full mxCell element including mxGeometry)
+- For delete: only cell_id is needed
+- Find the cell_id from "Current diagram XML" in system context
+- Example update: {"operations": [{"type": "update", "cell_id": "3", "new_xml": "<mxCell id=\\"3\\" value=\\"New Label\\" style=\\"rounded=1;\\" vertex=\\"1\\" parent=\\"1\\">\\n  <mxGeometry x=\\"100\\" y=\\"100\\" width=\\"120\\" height=\\"60\\" as=\\"geometry\\"/>\\n</mxCell>"}]}
+- Example delete: {"operations": [{"type": "delete", "cell_id": "5"}]}
+- Example add: {"operations": [{"type": "add", "cell_id": "new1", "new_xml": "<mxCell id=\\"new1\\" value=\\"New Box\\" style=\\"rounded=1;\\" vertex=\\"1\\" parent=\\"1\\">\\n  <mxGeometry x=\\"400\\" y=\\"200\\" width=\\"120\\" height=\\"60\\" as=\\"geometry\\"/>\\n</mxCell>"}]}
+
+⚠️ JSON ESCAPING: Every " inside new_xml MUST be escaped as \\". Example: id=\\"5\\" value=\\"Label\\"
 
 ## Draw.io XML Structure Reference
 
-Basic structure:
+**IMPORTANT:** You only generate the mxCell elements. The wrapper structure and root cells (id="0", id="1") are added automatically.
+
+Example - generate ONLY this:
 \`\`\`xml
-<mxGraphModel>
-  <root>
-    <mxCell id="0"/>
-    <mxCell id="1" parent="0"/>
-  </root>
-</mxGraphModel>
+<mxCell id="2" value="Label" style="rounded=1;" vertex="1" parent="1">
+  <mxGeometry x="100" y="100" width="120" height="60" as="geometry"/>
+</mxCell>
 \`\`\`
-Note: All other mxCell elements go as siblings after id="1".
 
 CRITICAL RULES:
-1. Always include the two root cells: <mxCell id="0"/> and <mxCell id="1" parent="0"/>
-2. ALL mxCell elements must be DIRECT children of <root> - NEVER nest mxCell inside another mxCell
-3. Use unique sequential IDs for all cells (start from "2" for user content)
-4. Set parent="1" for top-level shapes, or parent="<container-id>" for grouped elements
+1. Generate ONLY mxCell elements - NO wrapper tags (<mxfile>, <mxGraphModel>, <root>)
+2. Do NOT include root cells (id="0" or id="1") - they are added automatically
+3. ALL mxCell elements must be siblings - NEVER nest mxCell inside another mxCell
+4. Use unique sequential IDs starting from "2"
+5. Set parent="1" for top-level shapes, or parent="<container-id>" for grouped elements
 
 Shape (vertex) example:
 \`\`\`xml
@@ -113,15 +128,95 @@ Connector (edge) example:
 <mxCell id="3" style="endArrow=classic;html=1;" edge="1" parent="1" source="2" target="4">
   <mxGeometry relative="1" as="geometry"/>
 </mxCell>
+
+### Edge Routing Rules:
+When creating edges/connectors, you MUST follow these rules to avoid overlapping lines:
+
+**Rule 1: NEVER let multiple edges share the same path**
+- If two edges connect the same pair of nodes, they MUST exit/enter at DIFFERENT positions
+- Use exitY=0.3 for first edge, exitY=0.7 for second edge (NOT both 0.5)
+
+**Rule 2: For bidirectional connections (A↔B), use OPPOSITE sides**
+- A→B: exit from RIGHT side of A (exitX=1), enter LEFT side of B (entryX=0)
+- B→A: exit from LEFT side of B (exitX=0), enter RIGHT side of A (entryX=1)
+
+**Rule 3: Always specify exitX, exitY, entryX, entryY explicitly**
+- Every edge MUST have these 4 attributes set in the style
+- Example: style="edgeStyle=orthogonalEdgeStyle;exitX=1;exitY=0.3;entryX=0;entryY=0.3;endArrow=classic;"
+
+**Rule 4: Route edges AROUND intermediate shapes (obstacle avoidance) - CRITICAL!**
+- Before creating an edge, identify ALL shapes positioned between source and target
+- If any shape is in the direct path, you MUST use waypoints to route around it
+- For DIAGONAL connections: route along the PERIMETER (outside edge) of the diagram, NOT through the middle
+- Add 20-30px clearance from shape boundaries when calculating waypoint positions
+- Route ABOVE (lower y), BELOW (higher y), or to the SIDE of obstacles
+- NEVER draw a line that visually crosses over another shape's bounding box
+
+**Rule 5: Plan layout strategically BEFORE generating XML**
+- Organize shapes into visual layers/zones (columns or rows) based on diagram flow
+- Space shapes 150-200px apart to create clear routing channels for edges
+- Mentally trace each edge: "What shapes are between source and target?"
+- Prefer layouts where edges naturally flow in one direction (left-to-right or top-to-bottom)
+
+**Rule 6: Use multiple waypoints for complex routing**
+- One waypoint is often not enough - use 2-3 waypoints to create proper L-shaped or U-shaped paths
+- Each direction change needs a waypoint (corner point)
+- Waypoints should form clear horizontal/vertical segments (orthogonal routing)
+- Calculate positions by: (1) identify obstacle boundaries, (2) add 20-30px margin
+
+**Rule 7: Choose NATURAL connection points based on flow direction**
+- NEVER use corner connections (e.g., entryX=1,entryY=1) - they look unnatural
+- For TOP-TO-BOTTOM flow: exit from bottom (exitY=1), enter from top (entryY=0)
+- For LEFT-TO-RIGHT flow: exit from right (exitX=1), enter from left (entryX=0)
+- For DIAGONAL connections: use the side closest to the target, not corners
+- Example: Node below-right of source → exit from bottom (exitY=1) OR right (exitX=1), not corner
+
+**Before generating XML, mentally verify:**
+1. "Do any edges cross over shapes that aren't their source/target?" → If yes, add waypoints
+2. "Do any two edges share the same path?" → If yes, adjust exit/entry points
+3. "Are any connection points at corners (both X and Y are 0 or 1)?" → If yes, use edge centers instead
+4. "Could I rearrange shapes to reduce edge crossings?" → If yes, revise layout
+
+
 \`\`\`
 
+`
+
+// Style instructions - only included when minimalStyle is false
+const STYLE_INSTRUCTIONS = `
 Common styles:
 - Shapes: rounded=1 (rounded corners), fillColor=#hex, strokeColor=#hex
 - Edges: endArrow=classic/block/open/none, startArrow=none/classic, curved=1, edgeStyle=orthogonalEdgeStyle
 - Text: fontSize=14, fontStyle=1 (bold), align=center/left/right
 `
 
+// Minimal style instruction - skip styling and focus on layout (prepended to prompt for emphasis)
+const MINIMAL_STYLE_INSTRUCTION = `
+## ⚠️ MINIMAL STYLE MODE ACTIVE ⚠️
+
+### No Styling - Plain Black/White Only
+- NO fillColor, NO strokeColor, NO rounded, NO fontSize, NO fontStyle
+- NO color attributes (no hex colors like #ff69b4)
+- Style: "whiteSpace=wrap;html=1;" for shapes, "html=1;endArrow=classic;" for edges
+- IGNORE all color/style examples below
+
+### Container/Group Shapes - MUST be Transparent
+- For container shapes (boxes that contain other shapes): use "fillColor=none;" to make background transparent
+- This prevents containers from covering child elements
+- Example: style="whiteSpace=wrap;html=1;fillColor=none;" for container rectangles
+
+### Focus on Layout Quality
+Since we skip styling, STRICTLY follow the "Edge Routing Rules" section below:
+- SPACING: Minimum 50px gap between all elements
+- NO OVERLAPS: Elements and edges must never overlap
+- Follow ALL 7 Edge Routing Rules for arrow positioning
+- Use waypoints to route edges AROUND obstacles
+- Use different exitY/entryY values for multiple edges between same nodes
+
+`
+
 // Extended additions (~2600 tokens) - appended for models with 4000 token cache minimum
+// Total EXTENDED_SYSTEM_PROMPT = ~4400 tokens
 const EXTENDED_ADDITIONS = `
 
 ## Extended Tool Reference
@@ -129,172 +224,128 @@ const EXTENDED_ADDITIONS = `
 ### display_diagram Details
 
 **VALIDATION RULES** (XML will be rejected if violated):
-1. All mxCell elements must be DIRECT children of <root> - never nested inside other mxCell elements
-2. Every mxCell needs a unique id attribute
-3. Every mxCell (except id="0") needs a valid parent attribute referencing an existing cell
-4. Edge source/target attributes must reference existing cell IDs
-5. Escape special characters in values: &lt; for <, &gt; for >, &amp; for &, &quot; for "
-6. Always start with the two root cells: <mxCell id="0"/><mxCell id="1" parent="0"/>
+1. Generate ONLY mxCell elements - wrapper tags and root cells are added automatically
+2. All mxCell elements must be siblings - never nested inside other mxCell elements
+3. Every mxCell needs a unique id attribute (start from "2")
+4. Every mxCell needs a valid parent attribute (use "1" for top-level, or container-id for grouped)
+5. Edge source/target attributes must reference existing cell IDs
+6. Escape special characters in values: &lt; for <, &gt; for >, &amp; for &, &quot; for "
 
-**Example with swimlanes and edges** (note: all mxCells are siblings under <root>):
+**Example with swimlanes and edges** (generate ONLY this - no wrapper tags):
 \`\`\`xml
-<root>
-  <mxCell id="0"/>
-  <mxCell id="1" parent="0"/>
-  <mxCell id="lane1" value="Frontend" style="swimlane;" vertex="1" parent="1">
-    <mxGeometry x="40" y="40" width="200" height="200" as="geometry"/>
-  </mxCell>
-  <mxCell id="step1" value="Step 1" style="rounded=1;" vertex="1" parent="lane1">
-    <mxGeometry x="20" y="60" width="160" height="40" as="geometry"/>
-  </mxCell>
-  <mxCell id="lane2" value="Backend" style="swimlane;" vertex="1" parent="1">
-    <mxGeometry x="280" y="40" width="200" height="200" as="geometry"/>
-  </mxCell>
-  <mxCell id="step2" value="Step 2" style="rounded=1;" vertex="1" parent="lane2">
-    <mxGeometry x="20" y="60" width="160" height="40" as="geometry"/>
-  </mxCell>
-  <mxCell id="edge1" style="edgeStyle=orthogonalEdgeStyle;endArrow=classic;" edge="1" parent="1" source="step1" target="step2">
-    <mxGeometry relative="1" as="geometry"/>
-  </mxCell>
-</root>
+<mxCell id="lane1" value="Frontend" style="swimlane;" vertex="1" parent="1">
+  <mxGeometry x="40" y="40" width="200" height="200" as="geometry"/>
+</mxCell>
+<mxCell id="step1" value="Step 1" style="rounded=1;" vertex="1" parent="lane1">
+  <mxGeometry x="20" y="60" width="160" height="40" as="geometry"/>
+</mxCell>
+<mxCell id="lane2" value="Backend" style="swimlane;" vertex="1" parent="1">
+  <mxGeometry x="280" y="40" width="200" height="200" as="geometry"/>
+</mxCell>
+<mxCell id="step2" value="Step 2" style="rounded=1;" vertex="1" parent="lane2">
+  <mxGeometry x="20" y="60" width="160" height="40" as="geometry"/>
+</mxCell>
+<mxCell id="edge1" style="edgeStyle=orthogonalEdgeStyle;endArrow=classic;" edge="1" parent="1" source="step1" target="step2">
+  <mxGeometry relative="1" as="geometry"/>
+</mxCell>
 \`\`\`
 
+### append_diagram Details
+
+**WHEN TO USE:** Only call this tool when display_diagram output was truncated (you'll see an error message about truncation).
+
+**CRITICAL RULES:**
+1. Do NOT include any wrapper tags - just continue the mxCell elements
+2. Continue from EXACTLY where your previous output stopped
+3. Complete the remaining mxCell elements
+4. If still truncated, call append_diagram again with the next fragment
+
+**Example:** If previous output ended with \`<mxCell id="x" style="rounded=1\`, continue with \`;" vertex="1">...\` and complete the remaining elements.
+
 ### edit_diagram Details
 
-**CRITICAL RULES:**
-- Copy-paste the EXACT search pattern from the "Current diagram XML" in system context
-- Do NOT reorder attributes or reformat - the attribute order in draw.io XML varies and you MUST match it exactly
-- Only include the lines that are changing, plus 1-2 surrounding lines for context if needed
-- Break large changes into multiple smaller edits
-- Each search must contain complete lines (never truncate mid-line)
-- First match only - be specific enough to target the right element
+edit_diagram uses ID-based operations to modify cells directly by their id attribute.
+
+**Operations:**
+- **update**: Replace an existing cell. Provide cell_id and new_xml.
+- **add**: Add a new cell. Provide cell_id (new unique id) and new_xml.
+- **delete**: Remove a cell. Only cell_id is needed.
 
 **Input Format:**
 \`\`\`json
 {
-  "edits": [
-    {
-      "search": "EXACT lines copied from current XML (preserve attribute order!)",
-      "replace": "Replacement lines"
-    }
+  "operations": [
+    {"type": "update", "cell_id": "3", "new_xml": "<mxCell ...complete element...>"},
+    {"type": "add", "cell_id": "new1", "new_xml": "<mxCell ...new element...>"},
+    {"type": "delete", "cell_id": "5"}
   ]
 }
 \`\`\`
 
-## edit_diagram Best Practices
+**Examples:**
 
-### Core Principle: Unique & Precise Patterns
-Your search pattern MUST uniquely identify exactly ONE location in the XML. Before writing a search pattern:
-1. Review the "Current diagram XML" in the system context
-2. Identify the exact element(s) to modify by their unique id attribute
-3. Include enough context to ensure uniqueness
-
-### Pattern Construction Rules
-
-**Rule 1: Always include the element's id attribute**
+Change label:
 \`\`\`json
-{"search": "<mxCell id=\\"node5\\"", "replace": "<mxCell id=\\"node5\\" value=\\"New Label\\""}
+{"operations": [{"type": "update", "cell_id": "3", "new_xml": "<mxCell id=\\"3\\" value=\\"New Label\\" style=\\"rounded=1;\\" vertex=\\"1\\" parent=\\"1\\">\\n  <mxGeometry x=\\"100\\" y=\\"100\\" width=\\"120\\" height=\\"60\\" as=\\"geometry\\"/>\\n</mxCell>"}]}
 \`\`\`
 
-**Rule 2: Include complete XML elements when possible**
+Add new shape:
 \`\`\`json
-{
-  "search": "<mxCell id=\\"3\\" value=\\"Old\\" style=\\"rounded=1;\\" vertex=\\"1\\" parent=\\"1\\">\\n  <mxGeometry x=\\"100\\" y=\\"100\\" width=\\"120\\" height=\\"60\\" as=\\"geometry\\"/>\\n</mxCell>",
-  "replace": "<mxCell id=\\"3\\" value=\\"New\\" style=\\"rounded=1;\\" vertex=\\"1\\" parent=\\"1\\">\\n  <mxGeometry x=\\"100\\" y=\\"100\\" width=\\"120\\" height=\\"60\\" as=\\"geometry\\"/>\\n</mxCell>"
-}
+{"operations": [{"type": "add", "cell_id": "new1", "new_xml": "<mxCell id=\\"new1\\" value=\\"New Box\\" style=\\"rounded=1;fillColor=#dae8fc;\\" vertex=\\"1\\" parent=\\"1\\">\\n  <mxGeometry x=\\"400\\" y=\\"200\\" width=\\"120\\" height=\\"60\\" as=\\"geometry\\"/>\\n</mxCell>"}]}
 \`\`\`
 
-**Rule 3: Preserve exact whitespace and formatting**
-Copy the search pattern EXACTLY from the current XML, including leading spaces, line breaks (\\n), and attribute order.
+Delete cell:
+\`\`\`json
+{"operations": [{"type": "delete", "cell_id": "5"}]}
+\`\`\`
 
-### Good vs Bad Patterns
+**Error Recovery:**
+If cell_id not found, check "Current diagram XML" for correct IDs. Use display_diagram if major restructuring is needed
 
-**BAD:** \`{"search": "value=\\"Label\\""}\` - Too vague, matches multiple elements
-**BAD:** \`{"search": "<mxCell value=\\"X\\" id=\\"5\\""}\` - Reordered attributes won't match
-**GOOD:** \`{"search": "<mxCell id=\\"5\\" parent=\\"1\\" style=\\"...\\" value=\\"Old\\" vertex=\\"1\\">"}\` - Uses unique id with full context
 
-### Error Recovery
-If edit_diagram fails with "pattern not found":
-1. **First retry**: Check attribute order - copy EXACTLY from current XML
-2. **Second retry**: Expand context - include more surrounding lines
-3. **Third retry**: Try matching on just \`<mxCell id="X"\` prefix + full replacement
-4. **After 3 failures**: Fall back to display_diagram to regenerate entire diagram
 
-## Common Style Properties
 
-### Shape Styles
-- rounded=1, fillColor=#hex, strokeColor=#hex, strokeWidth=2
-- whiteSpace=wrap, html=1, opacity=50, shadow=1, glass=1
 
-### Edge/Connector Styles
-- endArrow=classic/block/open/oval/diamond/none, startArrow=none/classic
-- curved=1, edgeStyle=orthogonalEdgeStyle, strokeWidth=2
-- dashed=1, dashPattern=3 3, flowAnimation=1
+## Edge Examples
 
-### Text Styles
-- fontSize=14, fontStyle=1 (1=bold, 2=italic, 4=underline, 3=bold+italic)
-- fontColor=#hex, align=center/left/right, verticalAlign=middle/top/bottom
-
-## Common Shape Types
-
-### Basic Shapes
-- Rectangle: rounded=0;whiteSpace=wrap;html=1;
-- Rounded Rectangle: rounded=1;whiteSpace=wrap;html=1;
-- Ellipse/Circle: ellipse;whiteSpace=wrap;html=1;aspect=fixed;
-- Diamond: rhombus;whiteSpace=wrap;html=1;
-- Cylinder: shape=cylinder3;whiteSpace=wrap;html=1;
-
-### Flowchart Shapes
-- Process: rounded=1;whiteSpace=wrap;html=1;
-- Decision: rhombus;whiteSpace=wrap;html=1;
-- Start/End: ellipse;whiteSpace=wrap;html=1;
-- Document: shape=document;whiteSpace=wrap;html=1;
-- Database: shape=cylinder3;whiteSpace=wrap;html=1;
-
-### Container Types
-- Swimlane: swimlane;whiteSpace=wrap;html=1;
-- Group Box: rounded=1;whiteSpace=wrap;html=1;container=1;collapsible=0;
-
-## Container/Group Example
+### Two edges between same nodes (CORRECT - no overlap):
 \`\`\`xml
-<mxCell id="container1" value="Group Title" style="swimlane;whiteSpace=wrap;html=1;" vertex="1" parent="1">
-  <mxGeometry x="40" y="40" width="200" height="200" as="geometry"/>
+<mxCell id="e1" value="A to B" style="edgeStyle=orthogonalEdgeStyle;exitX=1;exitY=0.3;entryX=0;entryY=0.3;endArrow=classic;" edge="1" parent="1" source="a" target="b">
+  <mxGeometry relative="1" as="geometry"/>
 </mxCell>
-<mxCell id="child1" value="Child Element" style="rounded=1;" vertex="1" parent="container1">
-  <mxGeometry x="20" y="40" width="160" height="40" as="geometry"/>
+<mxCell id="e2" value="B to A" style="edgeStyle=orthogonalEdgeStyle;exitX=0;exitY=0.7;entryX=1;entryY=0.7;endArrow=classic;" edge="1" parent="1" source="b" target="a">
+  <mxGeometry relative="1" as="geometry"/>
 </mxCell>
 \`\`\`
 
-## Example: Complete Flowchart
-
+### Edge with single waypoint (simple detour):
 \`\`\`xml
-<root>
-  <mxCell id="0"/>
-  <mxCell id="1" parent="0"/>
-  <mxCell id="start" value="Start" style="ellipse;whiteSpace=wrap;html=1;fillColor=#d5e8d4;strokeColor=#82b366;" vertex="1" parent="1">
-    <mxGeometry x="200" y="40" width="100" height="60" as="geometry"/>
-  </mxCell>
-  <mxCell id="process1" value="Process Step" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#dae8fc;strokeColor=#6c8ebf;" vertex="1" parent="1">
-    <mxGeometry x="175" y="140" width="150" height="60" as="geometry"/>
-  </mxCell>
-  <mxCell id="decision" value="Decision?" style="rhombus;whiteSpace=wrap;html=1;fillColor=#fff2cc;strokeColor=#d6b656;" vertex="1" parent="1">
-    <mxGeometry x="175" y="240" width="150" height="100" as="geometry"/>
-  </mxCell>
-  <mxCell id="end" value="End" style="ellipse;whiteSpace=wrap;html=1;fillColor=#f8cecc;strokeColor=#b85450;" vertex="1" parent="1">
-    <mxGeometry x="200" y="380" width="100" height="60" as="geometry"/>
-  </mxCell>
-  <mxCell id="edge1" style="edgeStyle=orthogonalEdgeStyle;endArrow=classic;html=1;" edge="1" parent="1" source="start" target="process1">
-    <mxGeometry relative="1" as="geometry"/>
-  </mxCell>
-  <mxCell id="edge2" style="edgeStyle=orthogonalEdgeStyle;endArrow=classic;html=1;" edge="1" parent="1" source="process1" target="decision">
-    <mxGeometry relative="1" as="geometry"/>
-  </mxCell>
-  <mxCell id="edge3" value="Yes" style="edgeStyle=orthogonalEdgeStyle;endArrow=classic;html=1;" edge="1" parent="1" source="decision" target="end">
-    <mxGeometry relative="1" as="geometry"/>
-  </mxCell>
-</root>
+<mxCell id="edge1" style="edgeStyle=orthogonalEdgeStyle;exitX=0.5;exitY=1;entryX=0.5;entryY=0;endArrow=classic;" edge="1" parent="1" source="a" target="b">
+  <mxGeometry relative="1" as="geometry">
+    <Array as="points">
+      <mxPoint x="300" y="150"/>
+    </Array>
+  </mxGeometry>
+</mxCell>
 \`\`\`
-`
+
+### Edge with waypoints (routing AROUND obstacles) - CRITICAL PATTERN:
+**Scenario:** Hotfix(right,bottom) → Main(center,top), but Develop(center,middle) is in between.
+**WRONG:** Direct diagonal line crosses over Develop
+**CORRECT:** Route around the OUTSIDE (go right first, then up)
+\`\`\`xml
+<mxCell id="hotfix_to_main" style="edgeStyle=orthogonalEdgeStyle;exitX=0.5;exitY=0;entryX=1;entryY=0.5;endArrow=classic;" edge="1" parent="1" source="hotfix" target="main">
+  <mxGeometry relative="1" as="geometry">
+    <Array as="points">
+      <mxPoint x="750" y="80"/>
+      <mxPoint x="750" y="150"/>
+    </Array>
+  </mxGeometry>
+</mxCell>
+\`\`\`
+This routes the edge to the RIGHT of all shapes (x=750), then enters Main from the right side.
+
+**Key principle:** When connecting distant nodes diagonally, route along the PERIMETER of the diagram, not through the middle where other shapes exist.`
 
 // Extended system prompt = DEFAULT + EXTENDED_ADDITIONS
 export const EXTENDED_SYSTEM_PROMPT = DEFAULT_SYSTEM_PROMPT + EXTENDED_ADDITIONS
@@ -307,12 +358,16 @@ const EXTENDED_PROMPT_MODEL_PATTERNS = [
 ]
 
 /**
- * Get the appropriate system prompt based on the model ID
+ * Get the appropriate system prompt based on the model ID and style preference
  * Uses extended prompt for Opus 4.5 and Haiku 4.5 which have 4000 token cache minimum
  * @param modelId - The AI model ID from environment
+ * @param minimalStyle - If true, removes style instructions to save tokens
  * @returns The system prompt string
  */
-export function getSystemPrompt(modelId?: string): string {
+export function getSystemPrompt(
+    modelId?: string,
+    minimalStyle?: boolean,
+): string {
     const modelName = modelId || "AI"
 
     let prompt: string
@@ -333,5 +388,15 @@ export function getSystemPrompt(modelId?: string): string {
         prompt = DEFAULT_SYSTEM_PROMPT
     }
 
+    // Add style instructions based on preference
+    // Minimal style: prepend instruction at START (more prominent)
+    // Normal style: append at end
+    if (minimalStyle) {
+        console.log(`[System Prompt] Minimal style mode ENABLED`)
+        prompt = MINIMAL_STYLE_INSTRUCTION + prompt
+    } else {
+        prompt += STYLE_INSTRUCTIONS
+    }
+
     return prompt.replace("{{MODEL_NAME}}", modelName)
 }

lib/token-counter.ts

@@ -0,0 +1,39 @@
+/**
+ * Token counting utilities using js-tiktoken
+ *
+ * Uses cl100k_base encoding (GPT-4) which is close to Claude's tokenization.
+ * This is a pure JavaScript implementation, no WASM required.
+ */
+
+import { encodingForModel } from "js-tiktoken"
+import { DEFAULT_SYSTEM_PROMPT, EXTENDED_SYSTEM_PROMPT } from "./system-prompts"
+
+const encoder = encodingForModel("gpt-4o")
+
+/**
+ * Count the number of tokens in a text string
+ * @param text - The text to count tokens for
+ * @returns The number of tokens
+ */
+export function countTextTokens(text: string): number {
+    return encoder.encode(text).length
+}
+
+/**
+ * Get token counts for the system prompts
+ * Useful for debugging and optimizing prompt sizes
+ * @returns Object with token counts for default and extended prompts
+ */
+export function getSystemPromptTokenCounts(): {
+    default: number
+    extended: number
+    additions: number
+} {
+    const defaultTokens = countTextTokens(DEFAULT_SYSTEM_PROMPT)
+    const extendedTokens = countTextTokens(EXTENDED_SYSTEM_PROMPT)
+    return {
+        default: defaultTokens,
+        extended: extendedTokens,
+        additions: extendedTokens - defaultTokens,
+    }
+}

lib/use-file-processor.tsx

@@ -0,0 +1,110 @@
+"use client"
+
+import { useState } from "react"
+import { toast } from "sonner"
+import {
+    extractPdfText,
+    extractTextFileContent,
+    isPdfFile,
+    isTextFile,
+    MAX_EXTRACTED_CHARS,
+} from "@/lib/pdf-utils"
+
+export interface FileData {
+    text: string
+    charCount: number
+    isExtracting: boolean
+}
+
+/**
+ * Hook for processing file uploads, especially PDFs and text files.
+ * Handles text extraction, character limit validation, and cleanup.
+ */
+export function useFileProcessor() {
+    const [files, setFiles] = useState<File[]>([])
+    const [pdfData, setPdfData] = useState<Map<File, FileData>>(new Map())
+
+    const handleFileChange = async (newFiles: File[]) => {
+        setFiles(newFiles)
+
+        // Extract text immediately for new PDF/text files
+        for (const file of newFiles) {
+            const needsExtraction =
+                (isPdfFile(file) || isTextFile(file)) && !pdfData.has(file)
+            if (needsExtraction) {
+                // Mark as extracting
+                setPdfData((prev) => {
+                    const next = new Map(prev)
+                    next.set(file, {
+                        text: "",
+                        charCount: 0,
+                        isExtracting: true,
+                    })
+                    return next
+                })
+
+                // Extract text asynchronously
+                try {
+                    let text: string
+                    if (isPdfFile(file)) {
+                        text = await extractPdfText(file)
+                    } else {
+                        text = await extractTextFileContent(file)
+                    }
+
+                    // Check character limit
+                    if (text.length > MAX_EXTRACTED_CHARS) {
+                        const limitK = MAX_EXTRACTED_CHARS / 1000
+                        toast.error(
+                            `${file.name}: Content exceeds ${limitK}k character limit (${(text.length / 1000).toFixed(1)}k chars)`,
+                        )
+                        setPdfData((prev) => {
+                            const next = new Map(prev)
+                            next.delete(file)
+                            return next
+                        })
+                        // Remove the file from the list
+                        setFiles((prev) => prev.filter((f) => f !== file))
+                        continue
+                    }
+
+                    setPdfData((prev) => {
+                        const next = new Map(prev)
+                        next.set(file, {
+                            text,
+                            charCount: text.length,
+                            isExtracting: false,
+                        })
+                        return next
+                    })
+                } catch (error) {
+                    console.error("Failed to extract text:", error)
+                    toast.error(`Failed to read file: ${file.name}`)
+                    setPdfData((prev) => {
+                        const next = new Map(prev)
+                        next.delete(file)
+                        return next
+                    })
+                }
+            }
+        }
+
+        // Clean up pdfData for removed files
+        setPdfData((prev) => {
+            const next = new Map(prev)
+            for (const key of prev.keys()) {
+                if (!newFiles.includes(key)) {
+                    next.delete(key)
+                }
+            }
+            return next
+        })
+    }
+
+    return {
+        files,
+        pdfData,
+        handleFileChange,
+        setFiles, // Export for external control (e.g., clearing files)
+    }
+}

lib/use-quota-manager.tsx

@@ -0,0 +1,247 @@
+"use client"
+
+import { useCallback, useMemo } from "react"
+import { toast } from "sonner"
+import { QuotaLimitToast } from "@/components/quota-limit-toast"
+import { STORAGE_KEYS } from "@/lib/storage"
+
+export interface QuotaConfig {
+    dailyRequestLimit: number
+    dailyTokenLimit: number
+    tpmLimit: number
+}
+
+export interface QuotaCheckResult {
+    allowed: boolean
+    remaining: number
+    used: number
+}
+
+/**
+ * Hook for managing request/token quotas and rate limiting.
+ * Handles three types of limits:
+ * - Daily request limit
+ * - Daily token limit
+ * - Tokens per minute (TPM) rate limit
+ *
+ * Users with their own API key bypass all limits.
+ */
+export function useQuotaManager(config: QuotaConfig): {
+    hasOwnApiKey: () => boolean
+    checkDailyLimit: () => QuotaCheckResult
+    checkTokenLimit: () => QuotaCheckResult
+    checkTPMLimit: () => QuotaCheckResult
+    incrementRequestCount: () => void
+    incrementTokenCount: (tokens: number) => void
+    incrementTPMCount: (tokens: number) => void
+    showQuotaLimitToast: () => void
+    showTokenLimitToast: (used: number) => void
+    showTPMLimitToast: () => void
+} {
+    const { dailyRequestLimit, dailyTokenLimit, tpmLimit } = config
+
+    // Check if user has their own API key configured (bypass limits)
+    const hasOwnApiKey = useCallback((): boolean => {
+        const provider = localStorage.getItem(STORAGE_KEYS.aiProvider)
+        const apiKey = localStorage.getItem(STORAGE_KEYS.aiApiKey)
+        return !!(provider && apiKey)
+    }, [])
+
+    // Generic helper: Parse count from localStorage with NaN guard
+    const parseStorageCount = (key: string): number => {
+        const count = parseInt(localStorage.getItem(key) || "0", 10)
+        return Number.isNaN(count) ? 0 : count
+    }
+
+    // Generic helper: Create quota checker factory
+    const createQuotaChecker = useCallback(
+        (
+            getTimeKey: () => string,
+            timeStorageKey: string,
+            countStorageKey: string,
+            limit: number,
+        ) => {
+            return (): QuotaCheckResult => {
+                if (hasOwnApiKey())
+                    return { allowed: true, remaining: -1, used: 0 }
+                if (limit <= 0) return { allowed: true, remaining: -1, used: 0 }
+
+                const currentTime = getTimeKey()
+                const storedTime = localStorage.getItem(timeStorageKey)
+                let count = parseStorageCount(countStorageKey)
+
+                if (storedTime !== currentTime) {
+                    count = 0
+                    localStorage.setItem(timeStorageKey, currentTime)
+                    localStorage.setItem(countStorageKey, "0")
+                }
+
+                return {
+                    allowed: count < limit,
+                    remaining: limit - count,
+                    used: count,
+                }
+            }
+        },
+        [hasOwnApiKey],
+    )
+
+    // Generic helper: Create quota incrementer factory
+    const createQuotaIncrementer = useCallback(
+        (
+            getTimeKey: () => string,
+            timeStorageKey: string,
+            countStorageKey: string,
+            validateInput: boolean = false,
+        ) => {
+            return (tokens: number = 1): void => {
+                if (validateInput && (!Number.isFinite(tokens) || tokens <= 0))
+                    return
+
+                const currentTime = getTimeKey()
+                const storedTime = localStorage.getItem(timeStorageKey)
+                let count = parseStorageCount(countStorageKey)
+
+                if (storedTime !== currentTime) {
+                    count = 0
+                    localStorage.setItem(timeStorageKey, currentTime)
+                }
+
+                localStorage.setItem(countStorageKey, String(count + tokens))
+            }
+        },
+        [],
+    )
+
+    // Check daily request limit
+    const checkDailyLimit = useMemo(
+        () =>
+            createQuotaChecker(
+                () => new Date().toDateString(),
+                STORAGE_KEYS.requestDate,
+                STORAGE_KEYS.requestCount,
+                dailyRequestLimit,
+            ),
+        [createQuotaChecker, dailyRequestLimit],
+    )
+
+    // Increment request count
+    const incrementRequestCount = useMemo(
+        () =>
+            createQuotaIncrementer(
+                () => new Date().toDateString(),
+                STORAGE_KEYS.requestDate,
+                STORAGE_KEYS.requestCount,
+                false,
+            ),
+        [createQuotaIncrementer],
+    )
+
+    // Show quota limit toast (request-based)
+    const showQuotaLimitToast = useCallback(() => {
+        toast.custom(
+            (t) => (
+                <QuotaLimitToast
+                    used={dailyRequestLimit}
+                    limit={dailyRequestLimit}
+                    onDismiss={() => toast.dismiss(t)}
+                />
+            ),
+            { duration: 15000 },
+        )
+    }, [dailyRequestLimit])
+
+    // Check daily token limit
+    const checkTokenLimit = useMemo(
+        () =>
+            createQuotaChecker(
+                () => new Date().toDateString(),
+                STORAGE_KEYS.tokenDate,
+                STORAGE_KEYS.tokenCount,
+                dailyTokenLimit,
+            ),
+        [createQuotaChecker, dailyTokenLimit],
+    )
+
+    // Increment token count
+    const incrementTokenCount = useMemo(
+        () =>
+            createQuotaIncrementer(
+                () => new Date().toDateString(),
+                STORAGE_KEYS.tokenDate,
+                STORAGE_KEYS.tokenCount,
+                true, // Validate input tokens
+            ),
+        [createQuotaIncrementer],
+    )
+
+    // Show token limit toast
+    const showTokenLimitToast = useCallback(
+        (used: number) => {
+            toast.custom(
+                (t) => (
+                    <QuotaLimitToast
+                        type="token"
+                        used={used}
+                        limit={dailyTokenLimit}
+                        onDismiss={() => toast.dismiss(t)}
+                    />
+                ),
+                { duration: 15000 },
+            )
+        },
+        [dailyTokenLimit],
+    )
+
+    // Check TPM (tokens per minute) limit
+    const checkTPMLimit = useMemo(
+        () =>
+            createQuotaChecker(
+                () => Math.floor(Date.now() / 60000).toString(),
+                STORAGE_KEYS.tpmMinute,
+                STORAGE_KEYS.tpmCount,
+                tpmLimit,
+            ),
+        [createQuotaChecker, tpmLimit],
+    )
+
+    // Increment TPM count
+    const incrementTPMCount = useMemo(
+        () =>
+            createQuotaIncrementer(
+                () => Math.floor(Date.now() / 60000).toString(),
+                STORAGE_KEYS.tpmMinute,
+                STORAGE_KEYS.tpmCount,
+                true, // Validate input tokens
+            ),
+        [createQuotaIncrementer],
+    )
+
+    // Show TPM limit toast
+    const showTPMLimitToast = useCallback(() => {
+        const limitDisplay =
+            tpmLimit >= 1000 ? `${tpmLimit / 1000}k` : String(tpmLimit)
+        toast.error(
+            `Rate limit reached (${limitDisplay} tokens/min). Please wait 60 seconds before sending another request.`,
+            { duration: 8000 },
+        )
+    }, [tpmLimit])
+
+    return {
+        // Check functions
+        hasOwnApiKey,
+        checkDailyLimit,
+        checkTokenLimit,
+        checkTPMLimit,
+
+        // Increment functions
+        incrementRequestCount,
+        incrementTokenCount,
+        incrementTPMCount,
+
+        // Toast functions
+        showQuotaLimitToast,
+        showTokenLimitToast,
+        showTPMLimitToast,
+    }
+}

package.json

@@ -1,6 +1,6 @@
 {
     "name": "next-ai-draw-io",
-    "version": "0.2.0",
+    "version": "0.4.3",
     "license": "Apache-2.0",
     "private": true,
     "scripts": {
@@ -13,39 +13,47 @@
         "prepare": "husky"
     },
     "dependencies": {
-        "@ai-sdk/amazon-bedrock": "^3.0.62",
+        "@ai-sdk/amazon-bedrock": "^3.0.70",
         "@ai-sdk/anthropic": "^2.0.44",
         "@ai-sdk/azure": "^2.0.69",
         "@ai-sdk/deepseek": "^1.0.30",
+        "@ai-sdk/gateway": "^2.0.21",
         "@ai-sdk/google": "^2.0.0",
         "@ai-sdk/openai": "^2.0.19",
-        "@ai-sdk/react": "^2.0.22",
+        "@ai-sdk/react": "^2.0.107",
         "@aws-sdk/credential-providers": "^3.943.0",
         "@langfuse/client": "^4.4.9",
         "@langfuse/otel": "^4.4.4",
         "@langfuse/tracing": "^4.4.9",
         "@next/third-parties": "^16.0.6",
         "@openrouter/ai-sdk-provider": "^1.2.3",
+        "@opentelemetry/exporter-trace-otlp-http": "^0.208.0",
         "@opentelemetry/sdk-trace-node": "^2.2.0",
+        "@radix-ui/react-collapsible": "^1.1.12",
         "@radix-ui/react-dialog": "^1.1.6",
+        "@radix-ui/react-label": "^2.1.8",
         "@radix-ui/react-scroll-area": "^1.2.3",
         "@radix-ui/react-select": "^2.2.6",
         "@radix-ui/react-slot": "^1.1.2",
+        "@radix-ui/react-switch": "^1.2.6",
         "@radix-ui/react-tooltip": "^1.1.8",
-        "@vercel/analytics": "^1.5.0",
+        "@radix-ui/react-use-controllable-state": "^1.2.2",
         "@xmldom/xmldom": "^0.9.8",
         "ai": "^5.0.89",
         "base-64": "^1.0.0",
         "class-variance-authority": "^0.7.1",
         "clsx": "^2.1.1",
+        "js-tiktoken": "^1.0.21",
         "jsdom": "^26.0.0",
+        "jsonrepair": "^3.13.1",
         "lucide-react": "^0.483.0",
+        "motion": "^12.23.25",
         "next": "^16.0.7",
         "ollama-ai-provider-v2": "^1.5.4",
         "pako": "^2.1.0",
         "prism-react-renderer": "^2.4.1",
-        "react": "^19.0.0",
-        "react-dom": "^19.0.0",
+        "react": "^19.1.2",
+        "react-dom": "^19.1.2",
         "react-drawio": "^1.0.3",
         "react-icons": "^5.5.0",
         "react-markdown": "^10.1.0",
@@ -54,6 +62,7 @@
         "sonner": "^2.0.7",
         "tailwind-merge": "^3.0.2",
         "tailwindcss-animate": "^1.0.7",
+        "unpdf": "^1.4.0",
         "zod": "^4.1.12"
     },
     "lint-staged": {
@@ -63,6 +72,7 @@
         ]
     },
     "devDependencies": {
+        "@anthropic-ai/tokenizer": "^0.0.4",
         "@biomejs/biome": "2.3.8",
         "@tailwindcss/postcss": "^4",
         "@tailwindcss/typography": "^0.5.19",

packages/mcp-server/README.md

@@ -0,0 +1,162 @@
+# Next AI Draw.io MCP Server
+
+MCP (Model Context Protocol) server that enables AI agents like Claude Desktop and Cursor to generate and edit draw.io diagrams with **real-time browser preview**.
+
+**Self-contained** - includes an embedded HTTP server, no external dependencies required.
+
+## Quick Start
+
+```json
+{
+  "mcpServers": {
+    "drawio": {
+      "command": "npx",
+      "args": ["@next-ai-drawio/mcp-server@latest"]
+    }
+  }
+}
+```
+
+## Installation
+
+### Claude Desktop
+
+Add to your Claude Desktop config (`~/Library/Application Support/Claude/claude_desktop_config.json` on macOS):
+
+```json
+{
+  "mcpServers": {
+    "drawio": {
+      "command": "npx",
+      "args": ["@next-ai-drawio/mcp-server@latest"]
+    }
+  }
+}
+```
+
+### VS Code
+
+Add to your VS Code settings (`.vscode/mcp.json` in workspace or user settings):
+
+```json
+{
+  "mcpServers": {
+    "drawio": {
+      "command": "npx",
+      "args": ["@next-ai-drawio/mcp-server@latest"]
+    }
+  }
+}
+```
+
+### Cursor
+
+Add to Cursor MCP config (`~/.cursor/mcp.json`):
+
+```json
+{
+  "mcpServers": {
+    "drawio": {
+      "command": "npx",
+      "args": ["@next-ai-drawio/mcp-server@latest"]
+    }
+  }
+}
+```
+
+### Claude Code CLI
+
+```bash
+claude mcp add drawio -- npx @next-ai-drawio/mcp-server@latest
+```
+
+### Other MCP Clients
+
+Use the standard MCP configuration with:
+- **Command**: `npx`
+- **Args**: `["@next-ai-drawio/mcp-server@latest"]`
+
+## Usage
+
+1. Restart your MCP client after updating config
+2. Ask the AI to create a diagram:
+   > "Create a flowchart showing user authentication with login, MFA, and session management"
+3. The diagram appears in your browser in real-time!
+
+## Features
+
+- **Real-time Preview**: Diagrams appear and update in your browser as the AI creates them
+- **Natural Language**: Describe diagrams in plain text - flowcharts, architecture diagrams, etc.
+- **Edit Support**: Modify existing diagrams with natural language instructions
+- **Export**: Save diagrams as `.drawio` files
+- **Self-contained**: Embedded server, works offline (except draw.io UI which loads from embed.diagrams.net)
+
+## Available Tools
+
+| Tool | Description |
+|------|-------------|
+| `start_session` | Opens browser with real-time diagram preview |
+| `display_diagram` | Create a new diagram from XML |
+| `edit_diagram` | Edit diagram by ID-based operations (update/add/delete cells) |
+| `get_diagram` | Get the current diagram XML |
+| `export_diagram` | Save diagram to a `.drawio` file |
+
+## How It Works
+
+```
+┌─────────────────┐     stdio      ┌─────────────────┐
+│  Claude Desktop │ <───────────> │   MCP Server    │
+│    (AI Agent)   │               │  (this package) │
+└─────────────────┘               └────────┬────────┘
+                                          │
+                                 ┌────────▼────────┐
+                                 │ Embedded HTTP   │
+                                 │ Server (:6002)  │
+                                 └────────┬────────┘
+                                          │
+                                 ┌────────▼────────┐
+                                 │  User's Browser │
+                                 │ (draw.io embed) │
+                                 └─────────────────┘
+```
+
+1. **MCP Server** receives tool calls from Claude via stdio
+2. **Embedded HTTP Server** serves the draw.io UI and handles state
+3. **Browser** shows real-time diagram updates via polling
+
+## Configuration
+
+| Variable | Default | Description |
+|----------|---------|-------------|
+| `PORT` | `6002` | Port for the embedded HTTP server |
+
+## Troubleshooting
+
+### Port already in use
+
+If port 6002 is in use, the server will automatically try the next available port (up to 6020).
+
+Or set a custom port:
+```json
+{
+  "mcpServers": {
+    "drawio": {
+      "command": "npx",
+      "args": ["@next-ai-drawio/mcp-server@latest"],
+      "env": { "PORT": "6003" }
+    }
+  }
+}
+```
+
+### "No active session"
+
+Call `start_session` first to open the browser window.
+
+### Browser not updating
+
+Check that the browser URL has the `?mcp=` query parameter. The MCP session ID connects the browser to the server.
+
+## License
+
+Apache-2.0

packages/mcp-server/package.json

@@ -0,0 +1,55 @@
+{
+    "name": "@next-ai-drawio/mcp-server",
+    "version": "0.1.2",
+    "description": "MCP server for Next AI Draw.io - AI-powered diagram generation with real-time browser preview",
+    "type": "module",
+    "main": "dist/index.js",
+    "bin": {
+        "next-ai-drawio-mcp": "./dist/index.js"
+    },
+    "scripts": {
+        "build": "tsc",
+        "dev": "tsx watch src/index.ts",
+        "start": "node dist/index.js",
+        "prepublishOnly": "npm run build"
+    },
+    "keywords": [
+        "mcp",
+        "drawio",
+        "diagram",
+        "ai",
+        "claude",
+        "model-context-protocol"
+    ],
+    "author": "Biki-dev",
+    "license": "Apache-2.0",
+    "repository": {
+        "type": "git",
+        "url": "https://github.com/Biki-dev/next-ai-draw-io",
+        "directory": "packages/mcp-server"
+    },
+    "homepage": "https://next-ai-drawio.jiang.jp",
+    "bugs": {
+        "url": "https://github.com/Biki-dev/next-ai-draw-io/issues"
+    },
+    "publishConfig": {
+        "access": "public"
+    },
+    "dependencies": {
+        "@modelcontextprotocol/sdk": "^1.0.4",
+        "linkedom": "^0.18.0",
+        "open": "^10.1.0",
+        "zod": "^3.24.0"
+    },
+    "devDependencies": {
+        "@types/node": "^20",
+        "tsx": "^4.19.0",
+        "typescript": "^5"
+    },
+    "engines": {
+        "node": ">=18"
+    },
+    "files": [
+        "dist"
+    ]
+}

packages/mcp-server/src/diagram-operations.ts

@@ -0,0 +1,219 @@
+/**
+ * ID-based diagram operations
+ * Copied from lib/utils.ts to avoid cross-package imports
+ */
+
+export interface DiagramOperation {
+    type: "update" | "add" | "delete"
+    cell_id: string
+    new_xml?: string
+}
+
+export interface OperationError {
+    type: "update" | "add" | "delete"
+    cellId: string
+    message: string
+}
+
+export interface ApplyOperationsResult {
+    result: string
+    errors: OperationError[]
+}
+
+/**
+ * Apply diagram operations (update/add/delete) using ID-based lookup.
+ * This replaces the text-matching approach with direct DOM manipulation.
+ *
+ * @param xmlContent - The full mxfile XML content
+ * @param operations - Array of operations to apply
+ * @returns Object with result XML and any errors
+ */
+export function applyDiagramOperations(
+    xmlContent: string,
+    operations: DiagramOperation[],
+): ApplyOperationsResult {
+    const errors: OperationError[] = []
+
+    // Parse the XML
+    const parser = new DOMParser()
+    const doc = parser.parseFromString(xmlContent, "text/xml")
+
+    // Check for parse errors
+    const parseError = doc.querySelector("parsererror")
+    if (parseError) {
+        return {
+            result: xmlContent,
+            errors: [
+                {
+                    type: "update",
+                    cellId: "",
+                    message: `XML parse error: ${parseError.textContent}`,
+                },
+            ],
+        }
+    }
+
+    // Find the root element (inside mxGraphModel)
+    const root = doc.querySelector("root")
+    if (!root) {
+        return {
+            result: xmlContent,
+            errors: [
+                {
+                    type: "update",
+                    cellId: "",
+                    message: "Could not find <root> element in XML",
+                },
+            ],
+        }
+    }
+
+    // Build a map of cell IDs to elements
+    const cellMap = new Map<string, Element>()
+    root.querySelectorAll("mxCell").forEach((cell) => {
+        const id = cell.getAttribute("id")
+        if (id) cellMap.set(id, cell)
+    })
+
+    // Process each operation
+    for (const op of operations) {
+        if (op.type === "update") {
+            const existingCell = cellMap.get(op.cell_id)
+            if (!existingCell) {
+                errors.push({
+                    type: "update",
+                    cellId: op.cell_id,
+                    message: `Cell with id="${op.cell_id}" not found`,
+                })
+                continue
+            }
+
+            if (!op.new_xml) {
+                errors.push({
+                    type: "update",
+                    cellId: op.cell_id,
+                    message: "new_xml is required for update operation",
+                })
+                continue
+            }
+
+            // Parse the new XML
+            const newDoc = parser.parseFromString(
+                `<wrapper>${op.new_xml}</wrapper>`,
+                "text/xml",
+            )
+            const newCell = newDoc.querySelector("mxCell")
+            if (!newCell) {
+                errors.push({
+                    type: "update",
+                    cellId: op.cell_id,
+                    message: "new_xml must contain an mxCell element",
+                })
+                continue
+            }
+
+            // Validate ID matches
+            const newCellId = newCell.getAttribute("id")
+            if (newCellId !== op.cell_id) {
+                errors.push({
+                    type: "update",
+                    cellId: op.cell_id,
+                    message: `ID mismatch: cell_id is "${op.cell_id}" but new_xml has id="${newCellId}"`,
+                })
+                continue
+            }
+
+            // Import and replace the node
+            const importedNode = doc.importNode(newCell, true)
+            existingCell.parentNode?.replaceChild(importedNode, existingCell)
+
+            // Update the map with the new element
+            cellMap.set(op.cell_id, importedNode)
+        } else if (op.type === "add") {
+            // Check if ID already exists
+            if (cellMap.has(op.cell_id)) {
+                errors.push({
+                    type: "add",
+                    cellId: op.cell_id,
+                    message: `Cell with id="${op.cell_id}" already exists`,
+                })
+                continue
+            }
+
+            if (!op.new_xml) {
+                errors.push({
+                    type: "add",
+                    cellId: op.cell_id,
+                    message: "new_xml is required for add operation",
+                })
+                continue
+            }
+
+            // Parse the new XML
+            const newDoc = parser.parseFromString(
+                `<wrapper>${op.new_xml}</wrapper>`,
+                "text/xml",
+            )
+            const newCell = newDoc.querySelector("mxCell")
+            if (!newCell) {
+                errors.push({
+                    type: "add",
+                    cellId: op.cell_id,
+                    message: "new_xml must contain an mxCell element",
+                })
+                continue
+            }
+
+            // Validate ID matches
+            const newCellId = newCell.getAttribute("id")
+            if (newCellId !== op.cell_id) {
+                errors.push({
+                    type: "add",
+                    cellId: op.cell_id,
+                    message: `ID mismatch: cell_id is "${op.cell_id}" but new_xml has id="${newCellId}"`,
+                })
+                continue
+            }
+
+            // Import and append the node
+            const importedNode = doc.importNode(newCell, true)
+            root.appendChild(importedNode)
+
+            // Add to map
+            cellMap.set(op.cell_id, importedNode)
+        } else if (op.type === "delete") {
+            const existingCell = cellMap.get(op.cell_id)
+            if (!existingCell) {
+                errors.push({
+                    type: "delete",
+                    cellId: op.cell_id,
+                    message: `Cell with id="${op.cell_id}" not found`,
+                })
+                continue
+            }
+
+            // Check for edges referencing this cell (warning only, still delete)
+            const referencingEdges = root.querySelectorAll(
+                `mxCell[source="${op.cell_id}"], mxCell[target="${op.cell_id}"]`,
+            )
+            if (referencingEdges.length > 0) {
+                const edgeIds = Array.from(referencingEdges)
+                    .map((e) => e.getAttribute("id"))
+                    .join(", ")
+                console.warn(
+                    `[applyDiagramOperations] Deleting cell "${op.cell_id}" which is referenced by edges: ${edgeIds}`,
+                )
+            }
+
+            // Remove the node
+            existingCell.parentNode?.removeChild(existingCell)
+            cellMap.delete(op.cell_id)
+        }
+    }
+
+    // Serialize back to string
+    const serializer = new XMLSerializer()
+    const result = serializer.serializeToString(doc)
+
+    return { result, errors }
+}

packages/mcp-server/src/http-server.ts

@@ -0,0 +1,384 @@
+/**
+ * Embedded HTTP Server for MCP
+ *
+ * Serves a static HTML page with draw.io embed and handles state sync.
+ * This eliminates the need for an external Next.js app.
+ */
+
+import http from "node:http"
+import { log } from "./logger.js"
+
+interface SessionState {
+    xml: string
+    version: number
+    lastUpdated: Date
+}
+
+// In-memory state store (shared with MCP server in same process)
+export const stateStore = new Map<string, SessionState>()
+
+let server: http.Server | null = null
+let serverPort: number = 6002
+const MAX_PORT = 6020 // Don't retry beyond this port
+const SESSION_TTL = 60 * 60 * 1000 // 1 hour
+
+/**
+ * Get state for a session
+ */
+export function getState(sessionId: string): SessionState | undefined {
+    return stateStore.get(sessionId)
+}
+
+/**
+ * Set state for a session
+ */
+export function setState(sessionId: string, xml: string): number {
+    const existing = stateStore.get(sessionId)
+    const newVersion = (existing?.version || 0) + 1
+
+    stateStore.set(sessionId, {
+        xml,
+        version: newVersion,
+        lastUpdated: new Date(),
+    })
+
+    log.debug(`State updated: session=${sessionId}, version=${newVersion}`)
+    return newVersion
+}
+
+/**
+ * Start the embedded HTTP server
+ */
+export function startHttpServer(port: number = 6002): Promise<number> {
+    return new Promise((resolve, reject) => {
+        if (server) {
+            resolve(serverPort)
+            return
+        }
+
+        serverPort = port
+        server = http.createServer(handleRequest)
+
+        server.on("error", (err: NodeJS.ErrnoException) => {
+            if (err.code === "EADDRINUSE") {
+                if (port >= MAX_PORT) {
+                    reject(
+                        new Error(
+                            `No available ports in range 6002-${MAX_PORT}`,
+                        ),
+                    )
+                    return
+                }
+                log.info(`Port ${port} in use, trying ${port + 1}`)
+                server = null
+                startHttpServer(port + 1)
+                    .then(resolve)
+                    .catch(reject)
+            } else {
+                reject(err)
+            }
+        })
+
+        server.listen(port, () => {
+            serverPort = port
+            log.info(`Embedded HTTP server running on http://localhost:${port}`)
+            resolve(port)
+        })
+    })
+}
+
+/**
+ * Stop the HTTP server
+ */
+export function stopHttpServer(): void {
+    if (server) {
+        server.close()
+        server = null
+    }
+}
+
+/**
+ * Clean up expired sessions
+ */
+function cleanupExpiredSessions(): void {
+    const now = Date.now()
+    for (const [sessionId, state] of stateStore) {
+        if (now - state.lastUpdated.getTime() > SESSION_TTL) {
+            stateStore.delete(sessionId)
+            log.info(`Cleaned up expired session: ${sessionId}`)
+        }
+    }
+}
+
+// Run cleanup every 5 minutes
+setInterval(cleanupExpiredSessions, 5 * 60 * 1000)
+
+/**
+ * Get the current server port
+ */
+export function getServerPort(): number {
+    return serverPort
+}
+
+/**
+ * Handle HTTP requests
+ */
+function handleRequest(
+    req: http.IncomingMessage,
+    res: http.ServerResponse,
+): void {
+    const url = new URL(req.url || "/", `http://localhost:${serverPort}`)
+
+    // CORS headers for local development
+    res.setHeader("Access-Control-Allow-Origin", "*")
+    res.setHeader("Access-Control-Allow-Methods", "GET, POST, OPTIONS")
+    res.setHeader("Access-Control-Allow-Headers", "Content-Type")
+
+    if (req.method === "OPTIONS") {
+        res.writeHead(204)
+        res.end()
+        return
+    }
+
+    // Route handling
+    if (url.pathname === "/" || url.pathname === "/index.html") {
+        serveHtml(req, res, url)
+    } else if (
+        url.pathname === "/api/state" ||
+        url.pathname === "/api/mcp/state"
+    ) {
+        handleStateApi(req, res, url)
+    } else if (
+        url.pathname === "/api/health" ||
+        url.pathname === "/api/mcp/health"
+    ) {
+        res.writeHead(200, { "Content-Type": "application/json" })
+        res.end(JSON.stringify({ status: "ok", mcp: true }))
+    } else {
+        res.writeHead(404)
+        res.end("Not Found")
+    }
+}
+
+/**
+ * Serve the HTML page with draw.io embed
+ */
+function serveHtml(
+    req: http.IncomingMessage,
+    res: http.ServerResponse,
+    url: URL,
+): void {
+    const sessionId = url.searchParams.get("mcp") || ""
+
+    res.writeHead(200, { "Content-Type": "text/html" })
+    res.end(getHtmlPage(sessionId))
+}
+
+/**
+ * Handle state API requests
+ */
+function handleStateApi(
+    req: http.IncomingMessage,
+    res: http.ServerResponse,
+    url: URL,
+): void {
+    if (req.method === "GET") {
+        const sessionId = url.searchParams.get("sessionId")
+        if (!sessionId) {
+            res.writeHead(400, { "Content-Type": "application/json" })
+            res.end(JSON.stringify({ error: "sessionId required" }))
+            return
+        }
+
+        const state = stateStore.get(sessionId)
+        res.writeHead(200, { "Content-Type": "application/json" })
+        res.end(
+            JSON.stringify({
+                xml: state?.xml || null,
+                version: state?.version || 0,
+                lastUpdated: state?.lastUpdated?.toISOString() || null,
+            }),
+        )
+    } else if (req.method === "POST") {
+        let body = ""
+        req.on("data", (chunk) => {
+            body += chunk
+        })
+        req.on("end", () => {
+            try {
+                const { sessionId, xml } = JSON.parse(body)
+                if (!sessionId) {
+                    res.writeHead(400, { "Content-Type": "application/json" })
+                    res.end(JSON.stringify({ error: "sessionId required" }))
+                    return
+                }
+
+                const version = setState(sessionId, xml)
+                res.writeHead(200, { "Content-Type": "application/json" })
+                res.end(JSON.stringify({ success: true, version }))
+            } catch {
+                res.writeHead(400, { "Content-Type": "application/json" })
+                res.end(JSON.stringify({ error: "Invalid JSON" }))
+            }
+        })
+    } else {
+        res.writeHead(405)
+        res.end("Method Not Allowed")
+    }
+}
+
+/**
+ * Generate the HTML page with draw.io embed
+ */
+function getHtmlPage(sessionId: string): string {
+    return `<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <title>Draw.io MCP - ${sessionId || "No Session"}</title>
+    <style>
+        * { margin: 0; padding: 0; box-sizing: border-box; }
+        html, body { width: 100%; height: 100%; overflow: hidden; }
+        #container { width: 100%; height: 100%; display: flex; flex-direction: column; }
+        #header {
+            padding: 8px 16px;
+            background: #1a1a2e;
+            color: #eee;
+            font-family: system-ui, sans-serif;
+            font-size: 14px;
+            display: flex;
+            justify-content: space-between;
+            align-items: center;
+        }
+        #header .session { color: #888; font-size: 12px; }
+        #header .status { font-size: 12px; }
+        #header .status.connected { color: #4ade80; }
+        #header .status.disconnected { color: #f87171; }
+        #drawio { flex: 1; border: none; }
+    </style>
+</head>
+<body>
+    <div id="container">
+        <div id="header">
+            <div>
+                <strong>Draw.io MCP</strong>
+                <span class="session">${sessionId ? `Session: ${sessionId}` : "No MCP session"}</span>
+            </div>
+            <div id="status" class="status disconnected">Connecting...</div>
+        </div>
+        <iframe id="drawio" src="https://embed.diagrams.net/?embed=1&proto=json&spin=1&libraries=1"></iframe>
+    </div>
+
+    <script>
+        const sessionId = "${sessionId}";
+        const iframe = document.getElementById('drawio');
+        const statusEl = document.getElementById('status');
+
+        let currentVersion = 0;
+        let isDrawioReady = false;
+        let pendingXml = null;
+        let lastLoadedXml = null;
+
+        // Listen for messages from draw.io
+        window.addEventListener('message', (event) => {
+            if (event.origin !== 'https://embed.diagrams.net') return;
+
+            try {
+                const msg = JSON.parse(event.data);
+                handleDrawioMessage(msg);
+            } catch (e) {
+                // Ignore non-JSON messages
+            }
+        });
+
+        function handleDrawioMessage(msg) {
+            if (msg.event === 'init') {
+                isDrawioReady = true;
+                statusEl.textContent = 'Ready';
+                statusEl.className = 'status connected';
+
+                // Load pending XML if any
+                if (pendingXml) {
+                    loadDiagram(pendingXml);
+                    pendingXml = null;
+                }
+            } else if (msg.event === 'save') {
+                // User saved - push to state
+                if (msg.xml && msg.xml !== lastLoadedXml) {
+                    pushState(msg.xml);
+                }
+            } else if (msg.event === 'export') {
+                // Export completed
+                if (msg.data) {
+                    pushState(msg.data);
+                }
+            } else if (msg.event === 'autosave') {
+                // Autosave - push to state
+                if (msg.xml && msg.xml !== lastLoadedXml) {
+                    pushState(msg.xml);
+                }
+            }
+        }
+
+        function loadDiagram(xml) {
+            if (!isDrawioReady) {
+                pendingXml = xml;
+                return;
+            }
+
+            lastLoadedXml = xml;
+            iframe.contentWindow.postMessage(JSON.stringify({
+                action: 'load',
+                xml: xml,
+                autosave: 1
+            }), '*');
+        }
+
+        async function pushState(xml) {
+            if (!sessionId) return;
+
+            try {
+                const response = await fetch('/api/state', {
+                    method: 'POST',
+                    headers: { 'Content-Type': 'application/json' },
+                    body: JSON.stringify({ sessionId, xml })
+                });
+
+                if (response.ok) {
+                    const result = await response.json();
+                    currentVersion = result.version;
+                    lastLoadedXml = xml;
+                }
+            } catch (e) {
+                console.error('Failed to push state:', e);
+            }
+        }
+
+        async function pollState() {
+            if (!sessionId) return;
+
+            try {
+                const response = await fetch('/api/state?sessionId=' + encodeURIComponent(sessionId));
+                if (!response.ok) return;
+
+                const state = await response.json();
+
+                if (state.version && state.version > currentVersion && state.xml) {
+                    currentVersion = state.version;
+                    loadDiagram(state.xml);
+                }
+            } catch (e) {
+                console.error('Failed to poll state:', e);
+            }
+        }
+
+        // Start polling if we have a session
+        if (sessionId) {
+            pollState();
+            setInterval(pollState, 2000);
+        }
+    </script>
+</body>
+</html>`
+}

packages/mcp-server/src/index.ts

@@ -0,0 +1,476 @@
+#!/usr/bin/env node
+/**
+ * MCP Server for Next AI Draw.io
+ *
+ * Enables AI agents (Claude Desktop, Cursor, etc.) to generate and edit
+ * draw.io diagrams with real-time browser preview.
+ *
+ * Uses an embedded HTTP server - no external dependencies required.
+ */
+
+// Setup DOM polyfill for Node.js (required for XML operations)
+import { DOMParser } from "linkedom"
+;(globalThis as any).DOMParser = DOMParser
+
+// Create XMLSerializer polyfill using outerHTML
+class XMLSerializerPolyfill {
+    serializeToString(node: any): string {
+        if (node.outerHTML !== undefined) {
+            return node.outerHTML
+        }
+        if (node.documentElement) {
+            return node.documentElement.outerHTML
+        }
+        return ""
+    }
+}
+;(globalThis as any).XMLSerializer = XMLSerializerPolyfill
+
+import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js"
+import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js"
+import open from "open"
+import { z } from "zod"
+import {
+    applyDiagramOperations,
+    type DiagramOperation,
+} from "./diagram-operations.js"
+import {
+    getServerPort,
+    getState,
+    setState,
+    startHttpServer,
+} from "./http-server.js"
+import { log } from "./logger.js"
+
+// Server configuration
+const config = {
+    port: parseInt(process.env.PORT || "6002"),
+}
+
+// Session state (single session for simplicity)
+let currentSession: {
+    id: string
+    xml: string
+    version: number
+} | null = null
+
+// Create MCP server
+const server = new McpServer({
+    name: "next-ai-drawio",
+    version: "0.1.2",
+})
+
+// Register prompt with workflow guidance
+server.prompt(
+    "diagram-workflow",
+    "Guidelines for creating and editing draw.io diagrams",
+    () => ({
+        messages: [
+            {
+                role: "user",
+                content: {
+                    type: "text",
+                    text: `# Draw.io Diagram Workflow Guidelines
+
+## Creating a New Diagram
+1. Call start_session to open the browser preview
+2. Use display_diagram with complete mxGraphModel XML to create a new diagram
+
+## Adding Elements to Existing Diagram
+1. Use edit_diagram with "add" operation
+2. Provide a unique cell_id and complete mxCell XML
+3. No need to call get_diagram first - the server fetches latest state automatically
+
+## Modifying or Deleting Existing Elements
+1. FIRST call get_diagram to see current cell IDs and structure
+2. THEN call edit_diagram with "update" or "delete" operations
+3. For update, provide the cell_id and complete new mxCell XML
+
+## Important Notes
+- display_diagram REPLACES the entire diagram - only use for new diagrams
+- edit_diagram PRESERVES user's manual changes (fetches browser state first)
+- Always use unique cell_ids when adding elements (e.g., "shape-1", "arrow-2")`,
+                },
+            },
+        ],
+    }),
+)
+
+// Tool: start_session
+server.registerTool(
+    "start_session",
+    {
+        description:
+            "Start a new diagram session and open the browser for real-time preview. " +
+            "Starts an embedded server and opens a browser window with draw.io. " +
+            "The browser will show diagram updates as they happen.",
+        inputSchema: {},
+    },
+    async () => {
+        try {
+            // Start embedded HTTP server
+            const port = await startHttpServer(config.port)
+
+            // Create session
+            const sessionId = `mcp-${Date.now().toString(36)}-${Math.random().toString(36).substring(2, 8)}`
+            currentSession = {
+                id: sessionId,
+                xml: "",
+                version: 0,
+            }
+
+            // Open browser
+            const browserUrl = `http://localhost:${port}?mcp=${sessionId}`
+            await open(browserUrl)
+
+            log.info(`Started session ${sessionId}, browser at ${browserUrl}`)
+
+            return {
+                content: [
+                    {
+                        type: "text",
+                        text: `Session started successfully!\n\nSession ID: ${sessionId}\nBrowser URL: ${browserUrl}\n\nThe browser will now show real-time diagram updates.`,
+                    },
+                ],
+            }
+        } catch (error) {
+            const message =
+                error instanceof Error ? error.message : String(error)
+            log.error("start_session failed:", message)
+            return {
+                content: [{ type: "text", text: `Error: ${message}` }],
+                isError: true,
+            }
+        }
+    },
+)
+
+// Tool: display_diagram
+server.registerTool(
+    "display_diagram",
+    {
+        description:
+            "Display a NEW draw.io diagram from XML. REPLACES the entire diagram. " +
+            "Use this for creating new diagrams from scratch. " +
+            "To ADD elements to an existing diagram, use edit_diagram with 'add' operation instead. " +
+            "You should generate valid draw.io/mxGraph XML format.",
+        inputSchema: {
+            xml: z
+                .string()
+                .describe("The draw.io XML to display (mxGraphModel format)"),
+        },
+    },
+    async ({ xml }) => {
+        try {
+            if (!currentSession) {
+                return {
+                    content: [
+                        {
+                            type: "text",
+                            text: "Error: No active session. Please call start_session first.",
+                        },
+                    ],
+                    isError: true,
+                }
+            }
+
+            log.info(`Displaying diagram, ${xml.length} chars`)
+
+            // Update session state
+            currentSession.xml = xml
+            currentSession.version++
+
+            // Push to embedded server state
+            setState(currentSession.id, xml)
+
+            log.info(`Diagram displayed successfully`)
+
+            return {
+                content: [
+                    {
+                        type: "text",
+                        text: `Diagram displayed successfully!\n\nThe diagram is now visible in your browser.\n\nXML length: ${xml.length} characters`,
+                    },
+                ],
+            }
+        } catch (error) {
+            const message =
+                error instanceof Error ? error.message : String(error)
+            log.error("display_diagram failed:", message)
+            return {
+                content: [{ type: "text", text: `Error: ${message}` }],
+                isError: true,
+            }
+        }
+    },
+)
+
+// Tool: edit_diagram
+server.registerTool(
+    "edit_diagram",
+    {
+        description:
+            "Edit the current diagram by ID-based operations (update/add/delete cells). " +
+            "ALWAYS fetches the latest state from browser first, so user's manual changes are preserved.\n\n" +
+            "IMPORTANT workflow:\n" +
+            "- For ADD operations: Can use directly - just provide new unique cell_id and new_xml.\n" +
+            "- For UPDATE/DELETE: Call get_diagram FIRST to see current cell IDs, then edit.\n\n" +
+            "Operations:\n" +
+            "- add: Add a new cell. Provide cell_id (new unique id) and new_xml.\n" +
+            "- update: Replace an existing cell by its id. Provide cell_id and complete new_xml.\n" +
+            "- delete: Remove a cell by its id. Only cell_id is needed.\n\n" +
+            "For add/update, new_xml must be a complete mxCell element including mxGeometry.",
+        inputSchema: {
+            operations: z
+                .array(
+                    z.object({
+                        type: z
+                            .enum(["update", "add", "delete"])
+                            .describe("Operation type"),
+                        cell_id: z.string().describe("The id of the mxCell"),
+                        new_xml: z
+                            .string()
+                            .optional()
+                            .describe(
+                                "Complete mxCell XML element (required for update/add)",
+                            ),
+                    }),
+                )
+                .describe("Array of operations to apply"),
+        },
+    },
+    async ({ operations }) => {
+        try {
+            if (!currentSession) {
+                return {
+                    content: [
+                        {
+                            type: "text",
+                            text: "Error: No active session. Please call start_session first.",
+                        },
+                    ],
+                    isError: true,
+                }
+            }
+
+            // Fetch latest state from browser
+            const browserState = getState(currentSession.id)
+            if (browserState?.xml) {
+                currentSession.xml = browserState.xml
+                log.info("Fetched latest diagram state from browser")
+            }
+
+            if (!currentSession.xml) {
+                return {
+                    content: [
+                        {
+                            type: "text",
+                            text: "Error: No diagram to edit. Please create a diagram first with display_diagram.",
+                        },
+                    ],
+                    isError: true,
+                }
+            }
+
+            log.info(`Editing diagram with ${operations.length} operation(s)`)
+
+            // Apply operations
+            const { result, errors } = applyDiagramOperations(
+                currentSession.xml,
+                operations as DiagramOperation[],
+            )
+
+            if (errors.length > 0) {
+                const errorMessages = errors
+                    .map((e) => `${e.type} ${e.cellId}: ${e.message}`)
+                    .join("\n")
+                log.warn(`Edit had ${errors.length} error(s): ${errorMessages}`)
+            }
+
+            // Update state
+            currentSession.xml = result
+            currentSession.version++
+
+            // Push to embedded server
+            setState(currentSession.id, result)
+
+            log.info(`Diagram edited successfully`)
+
+            const successMsg = `Diagram edited successfully!\n\nApplied ${operations.length} operation(s).`
+            const errorMsg =
+                errors.length > 0
+                    ? `\n\nWarnings:\n${errors.map((e) => `- ${e.type} ${e.cellId}: ${e.message}`).join("\n")}`
+                    : ""
+
+            return {
+                content: [
+                    {
+                        type: "text",
+                        text: successMsg + errorMsg,
+                    },
+                ],
+            }
+        } catch (error) {
+            const message =
+                error instanceof Error ? error.message : String(error)
+            log.error("edit_diagram failed:", message)
+            return {
+                content: [{ type: "text", text: `Error: ${message}` }],
+                isError: true,
+            }
+        }
+    },
+)
+
+// Tool: get_diagram
+server.registerTool(
+    "get_diagram",
+    {
+        description:
+            "Get the current diagram XML (fetches latest from browser, including user's manual edits). " +
+            "Call this BEFORE edit_diagram if you need to update or delete existing elements, " +
+            "so you can see the current cell IDs and structure.",
+    },
+    async () => {
+        try {
+            if (!currentSession) {
+                return {
+                    content: [
+                        {
+                            type: "text",
+                            text: "Error: No active session. Please call start_session first.",
+                        },
+                    ],
+                    isError: true,
+                }
+            }
+
+            // Fetch latest state from browser
+            const browserState = getState(currentSession.id)
+            if (browserState?.xml) {
+                currentSession.xml = browserState.xml
+            }
+
+            if (!currentSession.xml) {
+                return {
+                    content: [
+                        {
+                            type: "text",
+                            text: "No diagram exists yet. Use display_diagram to create one.",
+                        },
+                    ],
+                }
+            }
+
+            return {
+                content: [
+                    {
+                        type: "text",
+                        text: `Current diagram XML:\n\n${currentSession.xml}`,
+                    },
+                ],
+            }
+        } catch (error) {
+            const message =
+                error instanceof Error ? error.message : String(error)
+            log.error("get_diagram failed:", message)
+            return {
+                content: [{ type: "text", text: `Error: ${message}` }],
+                isError: true,
+            }
+        }
+    },
+)
+
+// Tool: export_diagram
+server.registerTool(
+    "export_diagram",
+    {
+        description: "Export the current diagram to a .drawio file.",
+        inputSchema: {
+            path: z
+                .string()
+                .describe(
+                    "File path to save the diagram (e.g., ./diagram.drawio)",
+                ),
+        },
+    },
+    async ({ path }) => {
+        try {
+            if (!currentSession) {
+                return {
+                    content: [
+                        {
+                            type: "text",
+                            text: "Error: No active session. Please call start_session first.",
+                        },
+                    ],
+                    isError: true,
+                }
+            }
+
+            // Fetch latest state
+            const browserState = getState(currentSession.id)
+            if (browserState?.xml) {
+                currentSession.xml = browserState.xml
+            }
+
+            if (!currentSession.xml) {
+                return {
+                    content: [
+                        {
+                            type: "text",
+                            text: "Error: No diagram to export. Please create a diagram first.",
+                        },
+                    ],
+                    isError: true,
+                }
+            }
+
+            const fs = await import("node:fs/promises")
+            const nodePath = await import("node:path")
+
+            let filePath = path
+            if (!filePath.endsWith(".drawio")) {
+                filePath = `${filePath}.drawio`
+            }
+
+            const absolutePath = nodePath.resolve(filePath)
+            await fs.writeFile(absolutePath, currentSession.xml, "utf-8")
+
+            log.info(`Diagram exported to ${absolutePath}`)
+
+            return {
+                content: [
+                    {
+                        type: "text",
+                        text: `Diagram exported successfully!\n\nFile: ${absolutePath}\nSize: ${currentSession.xml.length} characters`,
+                    },
+                ],
+            }
+        } catch (error) {
+            const message =
+                error instanceof Error ? error.message : String(error)
+            log.error("export_diagram failed:", message)
+            return {
+                content: [{ type: "text", text: `Error: ${message}` }],
+                isError: true,
+            }
+        }
+    },
+)
+
+// Start the MCP server
+async function main() {
+    log.info("Starting MCP server for Next AI Draw.io (embedded mode)...")
+
+    const transport = new StdioServerTransport()
+    await server.connect(transport)
+
+    log.info("MCP server running on stdio")
+}
+
+main().catch((error) => {
+    log.error("Fatal error:", error)
+    process.exit(1)
+})

packages/mcp-server/src/logger.ts

@@ -0,0 +1,24 @@
+/**
+ * Logger for MCP server
+ *
+ * CRITICAL: MCP servers communicate via STDIO (stdin/stdout).
+ * Using console.log() will corrupt the JSON-RPC protocol messages.
+ * ALL logging MUST use console.error() which writes to stderr.
+ */
+
+export const log = {
+    info: (msg: string, ...args: unknown[]) => {
+        console.error(`[MCP-DrawIO] [INFO] ${msg}`, ...args)
+    },
+    error: (msg: string, ...args: unknown[]) => {
+        console.error(`[MCP-DrawIO] [ERROR] ${msg}`, ...args)
+    },
+    debug: (msg: string, ...args: unknown[]) => {
+        if (process.env.DEBUG === "true") {
+            console.error(`[MCP-DrawIO] [DEBUG] ${msg}`, ...args)
+        }
+    },
+    warn: (msg: string, ...args: unknown[]) => {
+        console.error(`[MCP-DrawIO] [WARN] ${msg}`, ...args)
+    },
+}

packages/mcp-server/tsconfig.json

@@ -0,0 +1,19 @@
+{
+    "compilerOptions": {
+        "target": "ES2022",
+        "module": "Node16",
+        "moduleResolution": "Node16",
+        "outDir": "./dist",
+        "rootDir": "./src",
+        "strict": true,
+        "esModuleInterop": true,
+        "skipLibCheck": true,
+        "forceConsistentCasingInFileNames": true,
+        "declaration": true,
+        "declarationMap": true,
+        "sourceMap": true,
+        "resolveJsonModule": true
+    },
+    "include": ["src/**/*"],
+    "exclude": ["node_modules", "dist"]
+}

public/chain-of-thought.txt

@@ -0,0 +1,65 @@
+Here is an extended summary of the paper **"Chain-of-Thought Prompting Elicits Reasoning in Large Language Models"** by Jason Wei, et al. This detailed overview covers the background, methodology, extensive experimental results, emergent properties, and qualitative analysis found in the study.
+
+### **1. Introduction and Motivation**
+The paper addresses a significant limitation in Large Language Models (LLMs): while scaling up model size (increasing parameters) has revolutionized performance on standard NLP tasks, it has not proven sufficient for challenging logical tasks such as arithmetic, commonsense, and symbolic reasoning.
+
+Traditional techniques to solve these problems fell into two camps:
+1.  **Finetuning:** Training models manually with large datasets of explanations (expensive and task-specific).
+2.  **Standard Few-Shot Prompting:** Providing input-output pairs (e.g., Question $\rightarrow$ Answer) without explaining *how* the answer was derived. This often fails on multi-step problems.
+
+The authors introduce **Chain-of-Thought (CoT) Prompting**, a simple method that combines the strengths of both approaches. It leverages the model's existing capabilities to generate natural language rationales without requiring any model parameter updates (finetuning).
+
+### **2. Methodology: What is Chain-of-Thought?**
+The core innovation is changing the structure of the "exemplars" (the few-shot examples included in the prompt).
+*   **Standard Prompting:** The model is shown a question and an immediate answer.
+    *   *Q: Roger has 5 balls. He buys 2 cans of 3 balls. How many now?*
+    *   *A: 11.*
+*   **Chain-of-Thought Prompting:** The model is shown a question, followed by a series of intermediate natural language reasoning steps that lead to the answer.
+    *   *A: Roger started with 5 balls. 2 cans of 3 tennis balls each is 6 tennis balls. 5 + 6 = 11. The answer is 11.*
+
+By interacting with the model using this format, the LLM learns to generate its own "thought process" for new, unseen questions. This allows the model to decompose complex problems into manageable intermediate steps.
+
+### **3. Experimental Setup**
+The researchers evaluated CoT prompting on several large language models, including **GPT-3 (175B)**, **LaMDA (137B)**, **PaLM (540B)**, **UL2 (20B)**, and **Codex**. They tested across three distinct domains of reasoning:
+*   **Arithmetic Reasoning:** Using benchmarks like **GSM8K** (math word problems), **SVAMP**, **ASDiv**, **AQuA**, and **MAWPS**.
+*   **Commonsense Reasoning:** Using datasets like **CSQA**, **StrategyQA**, **Date Understanding**, and **Sports Understanding**.
+*   **Symbolic Reasoning:** Using tasks like **Last Letter Concatenation** and **Coin Flip** tracking (determining if a coin is heads or tails after a sequence of flips).
+
+### **4. Key Findings and Results**
+
+#### **Arithmetic Reasoning**
+The results on math word problems were striking. Standard prompting struggled significantly, often exhibiting a flat scaling curve (performance didn't improve much even as models got bigger).
+*   **Performance Jump:** On the difficult **GSM8K** benchmark, **PaLM 540B** with CoT prompting achieved **56.9%** accuracy, compared to just 17.9% with standard prompting.
+*   **Surpassing State-of-the-Art:** PaLM 540B with CoT outperformed a previously finetuned GPT-3 model (55%), establishing a new state-of-the-art without needing a training set.
+*   **Calculator Integration:** The authors noted that some errors were simple calculation mistakes in otherwise correct logic. By hooking the CoT output into an external Python calculator, accuracy on GSM8K rose further to **58.6%**.
+
+#### **Commonsense Reasoning**
+CoT prompting improved performance on tasks requiring background knowledge and physical intuition.
+*   **StrategyQA:** PaLM 540B achieved **75.6%** accuracy via CoT, beating the prior state-of-the-art (69.4%).
+*   **Sports Understanding:** The model achieved **95.4%** accuracy, surpassing the performance of an unaided sports enthusiast (84%).
+*   The gains were minimal on CSQA, likely because many questions in that dataset did not require multi-step logic.
+
+#### **Symbolic Reasoning and Generalization**
+A unique strength of CoT was enabling **Out-of-Domain (OOD) Generalization**.
+*   In the **Coin Flip** task, the models were given examples with only 2 flips. However, using CoT, the models could successfully track coins flipped 3 or 4 times.
+*   Standard prompting failed completely on these longer sequences, while CoT allowed the model to repeat the logical steps as many times as necessary to reach the solution.
+
+### **5. Emergent Ability of Scale**
+One of the paper's most critical insights is that CoT reasoning is an **emergent ability** that depends on model size.
+*   **Small Models (<10B parameters):** CoT prompting provided **no benefit** and often hurt performance. Small models produced fluent but illogical chains of thought (hallucinations) or suffered from repetition.
+*   **Large Models (~100B+ parameters):** The ability to reason sequentially emerges at this scale. The performance gains from CoT are negligible for small models but increase dramatically for models like GPT-3 (175B) and PaLM (540B).
+
+### **6. Why Does It Work? (Ablation Studies)**
+To ensure the improvement was due to the reasoning steps and not other factors, the authors conducted three specific ablations:
+1.  **Equation Only:** They prompted the model to output just the math equation without words. This performed worse than CoT, suggesting that natural language helps the model "understand" the question semantics.
+2.  **Variable Compute:** They prompted the model to output dots (...) to consume compute time before answering. This yielded no improvement, proving that the *content* of the reasoning steps matters, not just the extra tokens.
+3.  **Reasoning After Answer:** They asked the model to give the answer first, then the explanation. This performed about the same as the baseline, proving that the chain of thought must come *before* the answer to guide the model's inference process.
+
+### **7. Error Analysis and Robustness**
+The authors manually analyzed errors made by the models.
+*   **Error Types:** In math problems, errors were categorized as **Semantic Understanding** (misunderstanding the question), **One-Step Missing** (skipping a logical step), or **Calculation Errors**.
+*   **Impact of Scale:** Scaling from PaLM 62B to PaLM 540B significantly reduced semantic and missing-step errors, confirming that larger models are better at logic, not just memorization.
+*   **Robustness:** The method proved robust to different annotators (different people writing the prompts) and different specific examples, though, like all prompting, different prompt styles did result in some variance.
+
+### **Conclusion**
+The paper establishes Chain-of-Thought prompting as a powerful paradigm for unlocking the reasoning potential of Large Language Models. By simply asking the model to "show its work," researchers can elicit complex logical behaviors that were previously thought to require specialized architectures or extensive finetuning. The work highlights that reasoning is an emergent capability of sufficiently large language models.

public/live-demo-button.svg

@@ -0,0 +1,4 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="140" height="36" viewBox="0 0 140 36">
+  <rect width="140" height="36" rx="8" fill="#6366f1"/>
+  <text x="70" y="24" font-family="-apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, sans-serif" font-size="15" font-weight="600" fill="white" text-anchor="middle">🚀 Live Demo</text>
+</svg>

scripts/test-diagram-operations.mjs

@@ -0,0 +1,251 @@
+/**
+ * Simple test script for applyDiagramOperations function
+ * Run with: node scripts/test-diagram-operations.mjs
+ */
+
+import { JSDOM } from "jsdom"
+
+// Set up DOMParser for Node.js environment
+const dom = new JSDOM()
+globalThis.DOMParser = dom.window.DOMParser
+globalThis.XMLSerializer = dom.window.XMLSerializer
+
+// Import the function (we'll inline it since it's not ESM exported)
+function applyDiagramOperations(xmlContent, operations) {
+    const errors = []
+    const parser = new DOMParser()
+    const doc = parser.parseFromString(xmlContent, "text/xml")
+
+    const parseError = doc.querySelector("parsererror")
+    if (parseError) {
+        return {
+            result: xmlContent,
+            errors: [{ type: "update", cellId: "", message: `XML parse error: ${parseError.textContent}` }],
+        }
+    }
+
+    const root = doc.querySelector("root")
+    if (!root) {
+        return {
+            result: xmlContent,
+            errors: [{ type: "update", cellId: "", message: "Could not find <root> element in XML" }],
+        }
+    }
+
+    const cellMap = new Map()
+    root.querySelectorAll("mxCell").forEach((cell) => {
+        const id = cell.getAttribute("id")
+        if (id) cellMap.set(id, cell)
+    })
+
+    for (const op of operations) {
+        if (op.type === "update") {
+            const existingCell = cellMap.get(op.cell_id)
+            if (!existingCell) {
+                errors.push({ type: "update", cellId: op.cell_id, message: `Cell with id="${op.cell_id}" not found` })
+                continue
+            }
+            if (!op.new_xml) {
+                errors.push({ type: "update", cellId: op.cell_id, message: "new_xml is required for update operation" })
+                continue
+            }
+            const newDoc = parser.parseFromString(`<wrapper>${op.new_xml}</wrapper>`, "text/xml")
+            const newCell = newDoc.querySelector("mxCell")
+            if (!newCell) {
+                errors.push({ type: "update", cellId: op.cell_id, message: "new_xml must contain an mxCell element" })
+                continue
+            }
+            const newCellId = newCell.getAttribute("id")
+            if (newCellId !== op.cell_id) {
+                errors.push({ type: "update", cellId: op.cell_id, message: `ID mismatch: cell_id is "${op.cell_id}" but new_xml has id="${newCellId}"` })
+                continue
+            }
+            const importedNode = doc.importNode(newCell, true)
+            existingCell.parentNode?.replaceChild(importedNode, existingCell)
+            cellMap.set(op.cell_id, importedNode)
+        } else if (op.type === "add") {
+            if (cellMap.has(op.cell_id)) {
+                errors.push({ type: "add", cellId: op.cell_id, message: `Cell with id="${op.cell_id}" already exists` })
+                continue
+            }
+            if (!op.new_xml) {
+                errors.push({ type: "add", cellId: op.cell_id, message: "new_xml is required for add operation" })
+                continue
+            }
+            const newDoc = parser.parseFromString(`<wrapper>${op.new_xml}</wrapper>`, "text/xml")
+            const newCell = newDoc.querySelector("mxCell")
+            if (!newCell) {
+                errors.push({ type: "add", cellId: op.cell_id, message: "new_xml must contain an mxCell element" })
+                continue
+            }
+            const newCellId = newCell.getAttribute("id")
+            if (newCellId !== op.cell_id) {
+                errors.push({ type: "add", cellId: op.cell_id, message: `ID mismatch: cell_id is "${op.cell_id}" but new_xml has id="${newCellId}"` })
+                continue
+            }
+            const importedNode = doc.importNode(newCell, true)
+            root.appendChild(importedNode)
+            cellMap.set(op.cell_id, importedNode)
+        } else if (op.type === "delete") {
+            const existingCell = cellMap.get(op.cell_id)
+            if (!existingCell) {
+                errors.push({ type: "delete", cellId: op.cell_id, message: `Cell with id="${op.cell_id}" not found` })
+                continue
+            }
+            existingCell.parentNode?.removeChild(existingCell)
+            cellMap.delete(op.cell_id)
+        }
+    }
+
+    const serializer = new XMLSerializer()
+    const result = serializer.serializeToString(doc)
+    return { result, errors }
+}
+
+// Test data
+const sampleXml = `<?xml version="1.0" encoding="UTF-8"?>
+<mxfile>
+  <diagram>
+    <mxGraphModel>
+      <root>
+        <mxCell id="0"/>
+        <mxCell id="1" parent="0"/>
+        <mxCell id="2" value="Box A" style="rounded=1;" vertex="1" parent="1">
+          <mxGeometry x="100" y="100" width="120" height="60" as="geometry"/>
+        </mxCell>
+        <mxCell id="3" value="Box B" style="rounded=1;" vertex="1" parent="1">
+          <mxGeometry x="300" y="100" width="120" height="60" as="geometry"/>
+        </mxCell>
+        <mxCell id="4" value="" style="edgeStyle=orthogonalEdgeStyle;" edge="1" parent="1" source="2" target="3">
+          <mxGeometry relative="1" as="geometry"/>
+        </mxCell>
+      </root>
+    </mxGraphModel>
+  </diagram>
+</mxfile>`
+
+let passed = 0
+let failed = 0
+
+function test(name, fn) {
+    try {
+        fn()
+        console.log(`✓ ${name}`)
+        passed++
+    } catch (e) {
+        console.log(`✗ ${name}`)
+        console.log(`  Error: ${e.message}`)
+        failed++
+    }
+}
+
+function assert(condition, message) {
+    if (!condition) throw new Error(message || "Assertion failed")
+}
+
+// Tests
+test("Update operation changes cell value", () => {
+    const { result, errors } = applyDiagramOperations(sampleXml, [
+        {
+            type: "update",
+            cell_id: "2",
+            new_xml: '<mxCell id="2" value="Updated Box A" style="rounded=1;" vertex="1" parent="1"><mxGeometry x="100" y="100" width="120" height="60" as="geometry"/></mxCell>',
+        },
+    ])
+    assert(errors.length === 0, `Expected no errors, got: ${JSON.stringify(errors)}`)
+    assert(result.includes('value="Updated Box A"'), "Updated value should be in result")
+    assert(!result.includes('value="Box A"'), "Old value should not be in result")
+})
+
+test("Update operation fails for non-existent cell", () => {
+    const { errors } = applyDiagramOperations(sampleXml, [
+        { type: "update", cell_id: "999", new_xml: '<mxCell id="999" value="Test"/>' },
+    ])
+    assert(errors.length === 1, "Should have one error")
+    assert(errors[0].message.includes("not found"), "Error should mention not found")
+})
+
+test("Update operation fails on ID mismatch", () => {
+    const { errors } = applyDiagramOperations(sampleXml, [
+        { type: "update", cell_id: "2", new_xml: '<mxCell id="WRONG" value="Test"/>' },
+    ])
+    assert(errors.length === 1, "Should have one error")
+    assert(errors[0].message.includes("ID mismatch"), "Error should mention ID mismatch")
+})
+
+test("Add operation creates new cell", () => {
+    const { result, errors } = applyDiagramOperations(sampleXml, [
+        {
+            type: "add",
+            cell_id: "new1",
+            new_xml: '<mxCell id="new1" value="New Box" style="rounded=1;" vertex="1" parent="1"><mxGeometry x="500" y="100" width="120" height="60" as="geometry"/></mxCell>',
+        },
+    ])
+    assert(errors.length === 0, `Expected no errors, got: ${JSON.stringify(errors)}`)
+    assert(result.includes('id="new1"'), "New cell should be in result")
+    assert(result.includes('value="New Box"'), "New cell value should be in result")
+})
+
+test("Add operation fails for duplicate ID", () => {
+    const { errors } = applyDiagramOperations(sampleXml, [
+        { type: "add", cell_id: "2", new_xml: '<mxCell id="2" value="Duplicate"/>' },
+    ])
+    assert(errors.length === 1, "Should have one error")
+    assert(errors[0].message.includes("already exists"), "Error should mention already exists")
+})
+
+test("Add operation fails on ID mismatch", () => {
+    const { errors } = applyDiagramOperations(sampleXml, [
+        { type: "add", cell_id: "new1", new_xml: '<mxCell id="WRONG" value="Test"/>' },
+    ])
+    assert(errors.length === 1, "Should have one error")
+    assert(errors[0].message.includes("ID mismatch"), "Error should mention ID mismatch")
+})
+
+test("Delete operation removes cell", () => {
+    const { result, errors } = applyDiagramOperations(sampleXml, [{ type: "delete", cell_id: "3" }])
+    assert(errors.length === 0, `Expected no errors, got: ${JSON.stringify(errors)}`)
+    assert(!result.includes('id="3"'), "Deleted cell should not be in result")
+    assert(result.includes('id="2"'), "Other cells should remain")
+})
+
+test("Delete operation fails for non-existent cell", () => {
+    const { errors } = applyDiagramOperations(sampleXml, [{ type: "delete", cell_id: "999" }])
+    assert(errors.length === 1, "Should have one error")
+    assert(errors[0].message.includes("not found"), "Error should mention not found")
+})
+
+test("Multiple operations in sequence", () => {
+    const { result, errors } = applyDiagramOperations(sampleXml, [
+        {
+            type: "update",
+            cell_id: "2",
+            new_xml: '<mxCell id="2" value="Updated" style="rounded=1;" vertex="1" parent="1"><mxGeometry x="100" y="100" width="120" height="60" as="geometry"/></mxCell>',
+        },
+        {
+            type: "add",
+            cell_id: "new1",
+            new_xml: '<mxCell id="new1" value="Added" style="rounded=1;" vertex="1" parent="1"><mxGeometry x="500" y="100" width="120" height="60" as="geometry"/></mxCell>',
+        },
+        { type: "delete", cell_id: "3" },
+    ])
+    assert(errors.length === 0, `Expected no errors, got: ${JSON.stringify(errors)}`)
+    assert(result.includes('value="Updated"'), "Updated value should be present")
+    assert(result.includes('id="new1"'), "Added cell should be present")
+    assert(!result.includes('id="3"'), "Deleted cell should not be present")
+})
+
+test("Invalid XML returns parse error", () => {
+    const { errors } = applyDiagramOperations("<not valid xml", [{ type: "delete", cell_id: "1" }])
+    assert(errors.length === 1, "Should have one error")
+})
+
+test("Missing root element returns error", () => {
+    const { errors } = applyDiagramOperations("<mxfile></mxfile>", [{ type: "delete", cell_id: "1" }])
+    assert(errors.length === 1, "Should have one error")
+    assert(errors[0].message.includes("root"), "Error should mention root element")
+})
+
+// Summary
+console.log(`\n${passed} passed, ${failed} failed`)
+process.exit(failed > 0 ? 1 : 0)

tsconfig.json

@@ -29,5 +29,5 @@
         ".next/types/**/*.ts",
         ".next/dev/types/**/*.ts"
     ],
-    "exclude": ["node_modules"]
+    "exclude": ["node_modules", "packages"]
 }

vercel.json

@@ -0,0 +1,12 @@
+{
+    "functions": {
+        "app/api/chat/route.ts": {
+            "memory": 512,
+            "maxDuration": 120
+        },
+        "app/api/**/route.ts": {
+            "memory": 256,
+            "maxDuration": 10
+        }
+    }
+}