Convert entire codebases across languages using multiple AI models and benchmark their performance.
| Feature | Description | |
|---|---|---|
| 🔁 | Multi-Model Conversion | Run the same conversion task across multiple AI models simultaneously |
| 📊 | Automatic Benchmark Scoring | Evaluate outputs using syntax validation, structural fidelity, and token efficiency |
| 🏆 | Leaderboard Rankings | See which AI models perform best across historical conversions |
| 🧠 | Language-Pair Benchmarking | Discover the best model for specific migrations like Python → Rust |
| 🧩 | Workspace Dashboard | Modern developer interface with model comparison and benchmark insights |
| 🔒 | Privacy-First Architecture | All conversions run directly from your browser to the AI provider — no backend |
| Python | JavaScript | TypeScript | Java | Go |
| Rust | C | C++ | C# | Ruby |
| PHP | Swift | Kotlin | Scala | Dart |
| R | Perl | Shell Script | Julia | MATLAB |
| Fortran | COBOL | Lisp |
CodexConvert works with any OpenAI-compatible API:
| Provider | Default Base URL |
|---|---|
| OpenAI | https://api.openai.com/v1 |
| DeepSeek | https://api.deepseek.com/v1 |
| Mistral | https://api.mistral.ai/v1 |
| Groq | https://api.groq.com/openai/v1 |
| Ollama (local) | http://localhost:11434/v1 |
| OpenRouter | https://openrouter.ai/api/v1 |
| Together AI | https://api.together.xyz/v1 |
| Custom | Any endpoint you trust |
┌──────────────────────┐
│ 📂 Upload Files │
│ (browser memory) │
└──────────┬───────────┘
│
▼
┌──────────────────────┐
│ 🔁 Model Execution │ Parallel dispatch
│ Manager │ (max 3 concurrent)
└──────────┬───────────┘
│
┌───────────┼───────────┐
▼ ▼ ▼
┌─────────┐ ┌─────────┐ ┌─────────┐
│ Model A │ │ Model B │ │ Model C │
│ (fetch) │ │ (fetch) │ │ (fetch) │
└────┬────┘ └────┬────┘ └────┬────┘
│ │ │
└───────────┼───────────┘
▼
┌──────────────────────┐
│ 📊 Benchmark Engine │ Syntax, structure,
│ + Scoring │ token efficiency
└──────────┬───────────┘
│
▼
┌──────────────────────┐
│ 💾 Benchmark Dataset│ localStorage
│ (max 200 runs) │ persistence
└──────────┬───────────┘
│
▼
┌──────────────────────┐
│ 🏆 Leaderboard │ Rankings, filtering,
│ System │ top models
└──────────────────────┘
| Layer | Responsibility |
|---|---|
| Model Execution Manager | Dispatches conversions in parallel with concurrency control and failure isolation |
| Benchmark Engine | Runs syntax check, structural fidelity, and token efficiency metrics on each output |
| Scoring | Combines metrics into a weighted 0–10 score per model |
| Benchmark Dataset | Persists runs to localStorage for historical analysis |
| Leaderboard System | Aggregates scores into global and language-pair rankings |
For the full architecture breakdown, see 📐 Architecture.
Every conversion is automatically evaluated using three metrics:
| Metric | Weight | What It Measures |
|---|---|---|
| ✅ Syntax Validity | 40% | Balanced brackets and delimiters |
| 🏗️ Structural Fidelity | 40% | Preservation of file count, paths, functions, classes, and imports |
| ⚡ Token Efficiency | 20% | Output conciseness relative to other models |
Scoring formula:
finalScore = syntaxScore × 0.4 + structuralScore × 0.4 + tokenScore × 0.2
Scores are normalized to a 0–10 scale. Full methodology: 📊 Benchmarking.
CodexConvert was built with a privacy-first architecture and has undergone a comprehensive security audit:
| Protection | |
|---|---|
| 🔐 | API keys stored in sessionStorage only — cleared when the tab closes |
| 🚫 | No backend server — zero server-side data collection |
| 📡 | Direct browser → AI provider communication |
| 🛡️ | Path sanitization prevents ZIP traversal attacks from LLM responses |
| 📦 | No user code stored in benchmark dataset — only scores and metadata |
| 🔍 | TypeScript strict mode enabled across the entire codebase |
| 🧱 | Content Security Policy restricts script sources and connections |
| 🔗 | HTTPS enforced for all remote provider URLs |
Full details: 🔒 Security.
.
├── App.tsx # Main app — state, conversion orchestration
├── index.tsx # React entry point
├── constants.ts # Languages, provider presets, model lists
├── types.ts # Shared TypeScript interfaces
│
├── core/
│ ├── modelExecutionManager.ts # Parallel model execution (max 3)
│ ├── benchmark/
│ │ ├── benchmarkEngine.ts # Runs metrics, produces results
│ │ ├── scoring.ts # Weighted scoring formula
│ │ ├── benchmarkDataset.ts # localStorage persistence (max 200 runs)
│ │ ├── rankingEngine.ts # Historical aggregation → rankings
│ │ ├── types.ts # Benchmark & leaderboard types
│ │ └── metrics/
│ │ ├── syntaxCheck.ts # Balanced bracket validation
│ │ ├── structuralFidelity.ts # File/path/element comparison
│ │ └── tokenUsage.ts # Token count estimation
│ └── leaderboard/
│ └── leaderboardEngine.ts # Global + language-pair queries
│
├── services/
│ └── llmService.ts # OpenAI-compatible API client + path sanitization
│
├── utils/
│ └── pathSanitizer.ts # ZIP path traversal prevention + display truncation
│
├── context/
│ ├── ProviderContext.tsx # LLM provider config (sessionStorage)
│ └── ToastContext.tsx # Notification system
│
├── components/
│ ├── layout/ # AppLayout, Sidebar, TopBar
│ ├── converter/ # ConversionWorkspace (3-panel)
│ ├── benchmark/ # ScorePanel (metrics breakdown)
│ ├── leaderboard/ # LeaderboardView, tables, widgets
│ ├── ProviderPicker.tsx # Provider/model configuration UI
│ ├── ModelSelector.tsx # Multi-model checkbox selector
│ ├── ComparisonPanel.tsx # Side-by-side model output cards
│ ├── CodeDisplay.tsx # Original ↔ converted viewer
│ ├── FileTree.tsx # Collapsible file tree
│ ├── PrivacyBadge.tsx # 🔒 Privacy mode indicator
│ └── Loader.tsx # Loading overlay
│
└── docs/
├── ARCHITECTURE.md # System architecture
├── BENCHMARKING.md # Evaluation methodology
└── SECURITY.md # Security model & privacy
| Phase | Feature | Status |
|---|---|---|
| Phase 1 | LLM Abstraction Layer | ✅ Complete |
| Phase 2 | Multi-Model Conversion | ✅ Complete |
| Phase 3 | Benchmark & Scoring Engine | ✅ Complete |
| Phase 4 | Open Benchmark Leaderboard | ✅ Complete |
| Phase 5 | Workspace UI Redesign | ✅ Complete |
| Phase 6 | Security Hardening & Documentation | ✅ Complete |
| Phase 7 | Community benchmark submissions & public leaderboard | 🔜 Planned |
| Phase 8 | Advanced metrics (AST comparison, runtime validation) | 🔜 Planned |
- Node.js >= 18
- An API key from any supported LLM provider
git clone https://github.com/aryanjsx/code-converter.git
cd code-converter
npm install
npm run devOpen the printed local URL (typically http://localhost:3000) in your browser.
- Select a provider from the dropdown (OpenAI, DeepSeek, Groq, etc.)
- Enter your API key
- Choose one or more models
- Upload a project folder or files
- Click Convert (or Compare N Models for multi-model)
No .env file is needed. All configuration happens in the browser at runtime.
| Layer | Technology |
|---|---|
| Framework | React 19 |
| Bundler | Vite 6 |
| Language | TypeScript 5.8 (strict mode) |
| AI Integration | Any OpenAI-compatible API |
| ZIP Export | JSZip + FileSaver |
| Styling | Tailwind CSS |
| Document | Description |
|---|---|
| 📐 Architecture | System design and subsystem breakdown |
| 📊 Benchmarking | Evaluation metrics and scoring methodology |
| 🔒 Security | Privacy model, audit results, and security practices |
| 🤝 Contributing | Development setup and contribution guide |
Contributions are welcome! Whether it's a new benchmark metric, a provider preset, a UI improvement, or a bug fix — check out our Contributing Guide for development setup, code structure, and pull request guidelines.
This project is licensed under the MIT License.