feat: Add GPT-5 model support with optimized parameters

.gitignore

@@ -172,3 +172,9 @@ cython_debug/
 
 # Project-specific files
 design/
+
+# macOS
+.DS_Store
+
+# Local translation files
+durex/

CLAUDE.md

@@ -1,17 +1,91 @@
-# Notes for Claude
+# CLAUDE.md
 
-## Environment Setup
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
 
-- Use `uv pip install` instead of regular pip for package installation
-- For local development installation: `uv pip install -e .`
+## Project Overview
 
-## Common Commands
+Tinbox is a CLI tool for translating large documents (especially PDFs) using LLMs. It handles model limitations, bypasses refusals, and uses multimodal capabilities for direct PDF translation without OCR.
 
-Use `tinbox --help` to see all available options. A few useful commands:
+## Development Commands
 
-- `tinbox --to es document.pdf` — translate a PDF to Spanish
-- `tinbox --from zh --to en document.docx` — translate a Word document from Chinese to English
+### Installation
+- Use `uv pip install -e .` for development installation (not regular pip)
+- Install with extras: `uv pip install -e ".[pdf,docx,dev]"`
 
-## Project-Specific Information
+### Testing
+```bash
+pytest                      # Run all tests
+pytest tests/test_cli.py   # Run specific test file
+pytest -k test_name        # Run specific test
+pytest --cov=tinbox        # Run with coverage
+```
 
-- Pillow is required for image processing in the PDF processor and LiteLLM translator
+### Code Quality
+```bash
+ruff check .               # Run linter
+ruff format .              # Format code
+mypy src/tinbox           # Type checking
+black src tests           # Alternative formatter
+isort src tests           # Sort imports
+```
+
+## Architecture
+
+### Translation Pipeline
+1. **Document Loading** (`core/processor/`): Handles PDF, DOCX, and text files
+   - PDF: Converts pages to images for multimodal processing
+   - DOCX: Extracts text while preserving structure
+   - Text: Direct text processing
+
+2. **Translation Algorithms** (`core/translation/algorithms.py`):
+   - **Page-by-page**: Default for PDFs, includes seam repair between pages
+   - **Sliding window**: For long text documents, maintains context overlap
+
+3. **Model Interfaces** (`core/translation/`):
+   - `interface.py`: Protocol definition for model implementations
+   - `litellm.py`: LiteLLM integration for OpenAI/Anthropic/Ollama
+   - Supports checkpointing for resumable translations
+
+4. **Cost Management** (`core/cost.py`):
+   - Estimates tokens and costs before translation
+   - Warns about high-cost operations
+   - Tracks actual usage during translation
+
+## Key Technical Details
+
+- **Model Specification Format**: `provider:model` (e.g., `anthropic:claude-3-sonnet`, `openai:gpt-4o`, `ollama:mistral-small`)
+  - **GPT-5 Support**: Full support for OpenAI's GPT-5 models (`gpt-5`, `gpt-5-mini`, `gpt-5-nano`)
+  - GPT-5 models automatically use optimized translation parameters:
+    - `reasoning_effort="low"` (faster processing, adequate quality for translation)
+    - `verbosity="low"` (concise output without explanations)
+    - Temperature/top_p are NOT supported by GPT-5 (automatically excluded)
+  - **GPT-5 Pricing** (accurate cost tracking implemented):
+    - `gpt-5`: $1.25/$10.00 per 1M tokens (input/output)
+    - `gpt-5-mini`: $0.25/$2.00 per 1M tokens (recommended for most translations)
+    - `gpt-5-nano`: $0.05/$0.40 per 1M tokens (fastest, most economical)
+- **Language codes**: ISO 639-1 format with aliases (e.g., 'en', 'zh', 'es')
+- **Source language 'auto'**: Accepted as valid value for automatic detection in LiteLLM translator
+- **Async architecture**: Core translation operations use asyncio
+- **Progress tracking**: Rich library for terminal UI
+- **Pillow dependency**: Required for PDF image processing
+
+## Common Usage Examples
+
+```bash
+# Basic translation
+tinbox translate --to es document.pdf
+
+# With specific model
+tinbox translate --model openai:gpt-4o --to fr document.docx
+
+# With GPT-5 models (optimized parameters applied automatically)
+tinbox translate --model openai:gpt-5 --to es document.pdf       # Highest quality
+tinbox translate --model openai:gpt-5-mini --to fr document.docx # Recommended
+tinbox translate --model openai:gpt-5-nano --to de document.txt  # Most economical
+
+# Cost estimation only
+tinbox translate --dry-run --to de large_document.pdf
+
+# Sliding window for long texts
+tinbox translate --algorithm sliding-window --to zh document.txt
+```

QUICKSTART_GUI.md

@@ -0,0 +1,149 @@
+# Tinbox GUI - Quick Start Guide
+
+Get the Tinbox Electron GUI up and running in 5 minutes!
+
+## Installation
+
+### Step 1: Install Python Dependencies
+
+```bash
+# Install Tinbox with GUI support
+uv pip install -e ".[gui,pdf,docx]"
+```
+
+This installs:
+- Tinbox core
+- FastAPI for the REST API server
+- Uvicorn for serving the API
+- PDF and DOCX support
+
+### Step 2: Install Node Dependencies
+
+```bash
+cd electron
+npm install
+```
+
+## Running the App
+
+### Development Mode
+
+From the `electron/` directory:
+
+```bash
+npm run dev
+```
+
+This will:
+1. Start the FastAPI backend on `http://127.0.0.1:8765`
+2. Start the Vite dev server on `http://localhost:5173`
+3. Launch the Electron app
+
+The app should open automatically with hot reload enabled.
+
+## First Use
+
+### 1. Configure API Keys
+
+Click the **settings icon** (⚙️) in the top right and enter your API keys:
+
+- **OpenAI**: For GPT-4, GPT-5 models
+- **Anthropic**: For Claude models
+- **Google**: For Gemini models (optional)
+
+### 2. Set Default Preferences
+
+In Settings, configure:
+- **Default Model**: e.g., `openai:gpt-4o-mini` (recommended for most users)
+- **Source Language**: `auto` for auto-detection
+- **Target Language**: e.g., `en` for English
+- **Algorithm**: `page-by-page` for PDFs (default)
+- **Output Format**: `text` (default)
+
+Click **Save Settings**.
+
+### 3. Translate Your First Document
+
+1. **Drag and drop** a PDF, DOCX, or TXT file into the window
+   - Or click the drop zone to browse for files
+
+2. **Review the cost estimate** that appears
+   - Shows estimated tokens and cost
+   - Click "OK" to proceed
+
+3. **Watch the progress** in real-time
+   - Progress bar shows current page
+   - Token count and cost update live
+   - Multiple files process in queue
+
+4. **Download the result** when complete
+   - Click the download icon (⬇️)
+   - Choose where to save the translation
+
+## Tips
+
+### Batch Processing
+- Drop multiple files at once
+- They'll queue automatically
+- Max 2 concurrent translations (configurable)
+
+### Cost Management
+- Start with `gpt-4o-mini` or `gpt-5-nano` for testing
+- Cost estimates help avoid surprises
+- View actual cost per job in the progress card
+
+### File Support
+- **PDF**: Converted to images, works with scanned docs
+- **DOCX**: Text extraction maintains structure
+- **TXT**: Direct text processing
+
+### Keyboard Shortcuts
+- **Cmd/Ctrl + ,**: Open settings (coming soon)
+- **Cmd/Ctrl + Q**: Quit app
+
+## Troubleshooting
+
+### "Failed to start Python server"
+
+**Solution**: Make sure you installed the GUI dependencies:
+```bash
+uv pip install -e ".[gui]"
+```
+
+### "API key not found"
+
+**Solution**: Either:
+1. Add key in Settings UI, or
+2. Set environment variable:
+   ```bash
+   export OPENAI_API_KEY="sk-..."
+   ```
+
+### Port 8765 already in use
+
+**Solution**:
+1. Stop other Tinbox instances
+2. Or change port in `electron/main.js` (line 13)
+
+### Translation stuck at 0%
+
+**Solution**:
+1. Check that your API key is valid
+2. Ensure you have sufficient API credits
+3. Check console for error messages
+
+## Next Steps
+
+- Explore different models and compare quality
+- Try both translation algorithms
+- Experiment with output formats
+- Check the full README for advanced features
+
+## Need Help?
+
+- Full documentation: `electron/README.md`
+- Python backend: `src/tinbox/api/`
+- React frontend: `electron/src/`
+- Report issues: https://github.com/anthropics/tinbox/issues
+
+Happy translating! 🎉

README.md

@@ -59,10 +59,38 @@ tinbox --to es document.pdf
   - Automatic context preservation between sections
 
 ### 🤖 Flexible Model Support
+- **GPT-5 Models**: Full support with optimized parameters (see below)
 - Use powerful cloud models (GPT-4V, Claude 3.5 Sonnet)
 - Run translations locally with Ollama
 - Mix and match models for different tasks
 
+#### GPT-5 Models (Recommended)
+
+Tinbox fully supports OpenAI's latest GPT-5 models with automatic parameter optimization:
+
+| Model | Best For | Cost (per 1M tokens) | Speed |
+|-------|----------|----------------------|-------|
+| **gpt-5** | Complex documents, highest quality | $1.25 / $10.00 (in/out) | Standard |
+| **gpt-5-mini** | Most translations (recommended) | $0.25 / $2.00 (in/out) | Fast |
+| **gpt-5-nano** | Simple docs, high volume | $0.05 / $0.40 (in/out) | Fastest |
+
+**Automatic Optimizations**:
+- `reasoning_effort="low"` - Faster processing for translation tasks
+- `verbosity="low"` - Concise output without unnecessary explanations
+- No temperature parameter (not supported by GPT-5)
+
+**Usage**:
+```bash
+# Recommended for most use cases
+tinbox --model openai:gpt-5-mini --to es document.pdf
+
+# For highest quality
+tinbox --model openai:gpt-5 --to fr complex_report.pdf
+
+# For maximum speed/economy
+tinbox --model openai:gpt-5-nano --to de simple_doc.txt
+```
+
 ### 🌐 Language Support
 - Flexible source/target language specification using ISO 639-1 codes
 - Common language aliases (e.g., "en", "zh", "es")

electron/.gitignore

@@ -0,0 +1,32 @@
+# Dependencies
+node_modules/
+package-lock.json
+
+# Build outputs
+dist/
+dist-electron/
+
+# Vite
+.vite/
+
+# TypeScript
+*.tsbuildinfo
+
+# OS files
+.DS_Store
+Thumbs.db
+
+# IDE
+.vscode/
+.idea/
+*.swp
+*.swo
+
+# Logs
+logs/
+*.log
+npm-debug.log*
+
+# Electron
+out/
+.webpack/