metadata
title: Rox AI
emoji: 🤖
colorFrom: indigo
colorTo: purple
sdk: docker
pinned: false
license: mit
app_port: 7860
Rox AI
A production-ready AI chat interface with multi-model support, file processing, real-time internet search, and seamless conversations.
Features
Ultra-Fast Performance
- Optimized streaming with instant chunk delivery (0ms flush interval)
- No rate limiting - unlimited API access
- Extended context windows (4x larger) for comprehensive understanding
- Parallel file processing for maximum speed
- Smart caching and request deduplication
- Ultra-fast retry logic (50-200ms)
Multi-Model Support
Choose from 7 powerful standalone LLM models:
| Model | Parameters | Specialty |
|---|---|---|
| Rox Core | 405B | Fast and efficient for everyday tasks |
| Rox 2.1 Turbo | 671B | Deep thinking and reasoning |
| Rox 3.5 Coder | 480B | Optimized for coding and development |
| Rox 4.5 Turbo | 685B | Advanced reasoning and analysis |
| Rox 5 Ultra | 14.8T datasets | Most powerful flagship model |
| Rox 6 Dyno | Latest Gen | Dynamic thinker with native vision |
| Rox 7 Coder | Latest Gen | Ultimate coding powerhouse with reasoning |
All Rox AI models are standalone, proprietary models developed from scratch by Rox AI Technologies.
Rox 6 Dyno - The Latest Innovation:
- Native multimodal vision built directly into the model architecture
- Extended context window (64k tokens) for long conversations
- Deep reasoning with transparent thinking process
- Superior at complex analysis and multi-step problem solving
- Processes images directly without requiring separate vision models
DeepResearch Mode (Available for Rox 5 Ultra, Rox 6 Dyno, and Rox 7 Coder)
DeepResearch is a premium research feature available for Rox 5 Ultra, Rox 6 Dyno, and Rox 7 Coder that provides comprehensive, in-depth analysis on any topic using real-time web data.
How to Enable:
- Select "Rox 5 Ultra", "Rox 6 Dyno", or "Rox 7 Coder" model from the model selector
- Toggle the "DeepResearch" switch in the input area
- Ask your question - the AI will conduct thorough research before responding
What DeepResearch Does:
- Executes 18+ search query variations across multiple search engines
- Reads up to 20 full articles for comprehensive understanding
- Analyzes 15+ different sources for accuracy
- Cross-references information across multiple sources
- Prioritizes the latest and most current information
Search Sources Used:
| Source | Type |
|---|---|
| SearXNG | Meta-search (Google, Bing, DuckDuckGo) |
| DuckDuckGo | Privacy-focused search |
| Wikipedia | Encyclopedia |
| Bing | Web search |
| arXiv | Research papers |
| GitHub | Code repositories |
| Community discussions | |
| Google News | Latest news |
| Hacker News | Tech discussions |
| StackOverflow | Programming Q&A |
| NPM/PyPI | Package registries |
| And more... | Specialized APIs |
Response Characteristics:
- Minimum 4500+ words for comprehensive coverage
- Structured with clear sections and headings
- Cites sources throughout the response
- Uses only numeric numbers (1, 2, 3) never Roman numerals
- Includes latest data from current year
- Covers all aspects: history, current state, future trends
DeepResearch Configuration:
| Setting | Value |
|---|---|
| Max Tokens | 32,768 |
| Temperature | 0.35 (focused) |
| Timeout | 15 minutes |
| Articles Read | Up to 20 |
| Search Variations | 18 queries |
| Min Sources | 15 |
Visual Indicators:
- Real-time status updates during research phase
- "DeepResearch" badge on responses
- Research statistics (searches performed, articles read)
- Badge preserved in PDF exports
Rox Vision - Integrated Image Understanding
Rox Vision is our dedicated vision-language model that powers image understanding across most Rox LLM models. It is seamlessly integrated and activates automatically when images are uploaded.
Vision Models:
| Model | Parameters | Role |
|---|---|---|
| Rox Vision | 90B | Primary vision model for image analysis |
| Rox Vision Max | Advanced | Backup model for enhanced reliability |
How It Works:
- User uploads an image with a question
- For Rox Core, 2.1 Turbo, 3.5 Coder, 4.5 Turbo, and 5 Ultra: Rox Vision automatically analyzes the image and extracts visual information
- For Rox 6 Dyno: Native vision processes images directly (no separate vision model needed)
- The analysis is passed to the selected LLM
- The LLM generates an intelligent response using the visual data
Capabilities:
- Scene analysis and composition understanding
- Object detection and identification
- Text extraction (OCR) from images and screenshots
- Visual reasoning and Q&A
- Support for JPG, PNG, GIF, WebP, and BMP formats
Note: Rox Vision is not a separately selectable model. It is integrated into Rox Core, 2.1 Turbo, 3.5 Coder, 4.5 Turbo, and 5 Ultra, and activates automatically when images are uploaded. Rox 6 Dyno has its own native vision capabilities built-in.
Screen Share - Voice-Controlled Screen Interaction (Desktop Only)
Screen Share is an exclusive desktop feature that combines screen capture, voice recognition, AI vision, and text-to-speech for a hands-free AI experience.
Key Features:
- 🎤 Voice Input: Speak your questions naturally - no typing needed
- 👁️ Visual Understanding: AI sees and analyzes your screen content
- 🔊 Voice Responses: AI reads answers aloud automatically
- 💬 Floating Window: Real-time display of prompts and responses
- 🤖 All Models: Choose from all 7 Rox AI models
How It Works:
- Click the Screen Share button (monitor icon in header)
- Select your preferred AI model
- Grant screen sharing and microphone permissions
- Start speaking - your words appear in real-time
- After 2.5 seconds of silence, AI automatically:
- Captures a screenshot
- Processes your question + screen content
- Responds with text and voice
Requirements:
- Desktop computer (1024px+ width)
- Chrome 94+, Edge 94+, or Safari 13+
- Microphone and screen sharing permissions
- Not available on mobile devices
Use Cases:
- 💻 Coding Help: Share code editor, ask for debugging help
- 📚 Learning: Share tutorials, get explanations
- 🔧 Troubleshooting: Share error messages, get solutions
- 🎨 Design Review: Share mockups, get feedback
- 📊 Data Analysis: Share charts, get insights
Documentation:
- User Guide - Complete instructions
- Technical Docs - Developer reference
- Quick Reference - Cheat sheet
Vision Processing:
- Rox 6 Dyno: Native vision - processes screenshots directly (fastest)
- Other Models: Use Rox Vision for image analysis (automatic)
Live Internet Search
- Real-time web search for latest news, events, and information
- Multiple search sources with intelligent fallback
- Automatic detection of queries requiring live data
- Visual indicator shows when responses use internet data
Specialized API Integrations
Rox AI includes several free API integrations that provide specialized data without requiring API keys:
| API | Purpose | Trigger Examples |
|---|---|---|
| Open-Meteo | Weather forecasts and conditions | "Weather in Tokyo", "Temperature in New York" |
| Currency API | Live exchange rates | "Dollar to rupee", "USD to INR", "Exchange rate" |
| CoinGecko | Cryptocurrency prices | "Bitcoin price", "ETH to USD", "Dogecoin today" |
| TheSportsDB | Live sports scores | "IPL score", "RCB vs CSK", "NBA results", "Premier League" |
| Yahoo Finance | Stock market data | "Nifty today", "Reliance stock price", "Sensex live" |
| Open Library | Book information and search | "Books by Stephen King", "Who wrote 1984" |
| arXiv | Research papers and academic studies | "Research on machine learning", "Latest papers on quantum computing" |
| IP-API | Geolocation from IP addresses | "What is my IP", "My location" |
| GitHub | Repository and code search | "GitHub repos for React", "Code libraries for Python" |
How Specialized APIs Work:
- User query is analyzed for specialized patterns
- If a pattern matches (weather, currency, crypto, sports, stocks, books, research, etc.), the appropriate API is called
- Results are formatted and returned directly for accurate, structured data
- If specialized API fails, system falls back to general web search
Real-Time Data (No Caching): Weather, Currency, Cryptocurrency, Sports, Stock, and IP queries always fetch fresh data - they are never cached to ensure accuracy.
Benefits:
- More accurate and structured data for specific query types
- Faster responses for specialized queries
- No API keys required (100% free services)
- Automatic fallback ensures reliability
File Processing
Upload and analyze documents:
- PDF parsing with text extraction
- Word documents (.docx) with full text extraction
- Excel spreadsheets (.xlsx)
- PowerPoint presentations (.pptx)
- RTF documents
- Code files (60+ languages)
- Text and data files (CSV, JSON, YAML, XML, etc.)
- Images with Rox Vision analysis
Modern UI/UX
- Dark/Light theme toggle
- Smooth animations
- Mobile-responsive design
- Code syntax highlighting
- Math rendering with KaTeX
Advanced Features
- DeepResearch mode for comprehensive analysis (Rox 5 Ultra, Rox 6 Dyno, and Rox 7 Coder)
- Screen Share (Desktop Only) - Share your screen and interact with AI using voice commands
- Conversation history with persistence
- Message editing and regeneration
- Text-to-speech for responses
- PDF export with DeepResearch badge
- Keyboard shortcuts
- PWA support (installable)
Deployment
Deploy to Hugging Face Spaces using Docker, or run locally:
# Local development
npm install
cp .env.example .env # Configure your NVIDIA API key
npm start
Environment Variables
| Variable | Required | Description |
|---|---|---|
NVIDIA_API_KEY |
Yes | Your NVIDIA API key from build.nvidia.com |
PORT |
No | Server port (default: 7860) |
HOST |
No | Server host (default: 0.0.0.0) |
NODE_ENV |
No | Environment mode (production/development) |
API Endpoints
| Endpoint | Method | Description |
|---|---|---|
/api/chat |
POST | Send messages to AI |
/api/health |
GET | Health check with system info |
/api/models |
GET | List available models |
/api/version |
GET | Get server version |
Security
- Input validation and sanitization
- XSS protection
- CORS configuration
- Rate limiting (1000 req/min)
- Security headers (CSP, HSTS, etc.)
- No sensitive data logging
- Non-root Docker user
License
MIT License
Built by Mohammad Faiz, CEO & Founder of Rox AI Technologies