A next-generation browser that combines intelligent web research with experimental AI-powered 3D world creation using Three.js. Built for the OpenAI Open Model Hackathon, leveraging gpt-oss models for advanced reasoning and 3D generation capabilities.
🏆 OpenAI Open Model Hackathon Submission - Categories: Best Overall, Best Local Agent, Wildcard
- Intelligent Web Browsing: AI-powered web research with multiple search strategies
- Deep Search Engine: Revolutionary "Open Tabs Tool" that autonomously opens and analyzes multiple web pages simultaneously
- Agent System: Autonomous AI agents for complex multi-step research tasks
- Universal Model Support:
- All open-source models (via Ollama, vLLM, Hugging Face)
- Major providers (OpenAI, Anthropic, Google, etc.)
- gpt-oss integration: Leveraging OpenAI's new open-weight reasoning models (120B & 20B)
- Enhanced Search Modes: Deep web search with comprehensive cross-reference analysis
- Contextual Research: AI maintains context across multiple sources and sessions
- Three.js Integration: Real-time 3D world rendering and manipulation in the browser
- AI-Powered 3D Generation: Experimental text-to-3D capabilities using LLMs and 3D AI providers
- Basic Shape Creation: Create and manipulate primitive shapes (boxes, spheres, cylinders)
- Future-Ready Architecture: Designed for advanced 3D generation as AI models improve
- World Instance Management: Multiple isolated 3D environments with resource optimization
- Real-time Rendering: 60 FPS browser-based 3D rendering with quality controls
- Deep Search "Open Tabs Tool":
- Autonomously opens multiple relevant web pages
- Cross-references information across sources
- Builds comprehensive knowledge graphs
- Maintains context across browsing sessions
- 3D Object Management: Add, remove, and modify 3D objects in real-time
- Lighting System: Dynamic Three.js lighting with directional, point, and spot lights
- Material System: PBR materials with customizable properties
- Camera Controls: Perspective and orthographic viewport modes
- Render Quality: Adjustable rendering quality (low, medium, high)
- Experimental AI 3D: Text-to-3D generation (expanding as AI models evolve)
- Best Overall: Revolutionary browser combining AI research with 3D creation
- Best Local Agent: Advanced offline AI research capabilities with local model support
- Wildcard: Unexpected fusion of web browsing and 3D world generation
OramBrowser showcases the power of OpenAI's new open-weight reasoning models:
- gpt-oss-120B & gpt-oss-20B: Advanced reasoning for complex research tasks
- Local Processing: Run entirely offline with local model instances
- Creative Applications: Novel use of reasoning models for 3D world understanding and generation
- Autonomous Web Research: AI agents that think, browse, and synthesize information
- Multi-Modal Intelligence: Bridging text research with spatial 3D creation
- Open Source Ecosystem: Supporting all open-source models and providers
- Future-Proof Architecture: Ready for next-generation AI capabilities
- Node.js 18+ and npm
- Python 3.8+ (for backend services)
- gpt-oss models (optional): Download from Hugging Face
- Local LLM providers (optional): Ollama, vLLM, or LM Studio for offline capabilities
-
Clone the repository
git clone https://github.com/your-username/OramBrowser.git cd OramBrowser -
Install dependencies
npm install cd electron/frontend npm install -
Configure AI Models
- Set up local models (Ollama recommended for gpt-oss)
- Configure API keys for cloud providers (optional)
- Enable experimental 3D AI generation features
-
Start the application
npm run dev
OramBrowser's revolutionary deep search engine automatically:
- Autonomous Browsing: AI opens multiple relevant web pages based on your query
- Cross-Reference Analysis: Compares information across sources for accuracy
- Context Preservation: Maintains research context across multiple browsing sessions
- Knowledge Synthesis: Combines findings into comprehensive research reports
- Enter your research query in the enhanced search bar
- Select "Deep Search" mode
- Watch as AI agents autonomously open and analyze relevant sources
- Review the synthesized findings and source verification
- Continue research with AI maintaining full context
- Open Source: All Hugging Face models, Ollama collections, local instances
- Cloud Providers: OpenAI (including gpt-oss), Anthropic, Google, Azure
- Reasoning Models: Specialized support for gpt-oss-120B and gpt-oss-20B
- Local Processing: Full offline capability with local model deployment
- Click the "+" button in the tab bar
- Select "3D World" from the dropdown
- A new Three.js World tab will open with an empty 3D scene
- Load Scene: Click "Load Scene" to create a basic environment with lighting
- Add Objects: Use the object controls to add boxes, spheres, and cylinders
- Add Lights: Add directional (sun) or point lights to illuminate your scene
- Clear World: Remove all objects to start fresh
- Text-to-3D (Beta): Describe objects in natural language for AI generation
- Scene Understanding: AI analyzes and suggests improvements to 3D scenes
- Future Enhancements: Expanding capabilities as 3D AI models improve
- Fullscreen Mode: Click the fullscreen button for immersive 3D exploration
- Render Stats: Monitor FPS, draw calls, and triangle count
- Quality Settings: Adjust rendering quality for performance vs. visual fidelity
- Multiple Worlds: Create multiple 3D worlds in separate tabs using Three.js
- Resource Management: Automatic memory management and instance isolation
- State Persistence: Worlds maintain their state when switching between tabs
- Auto-Suspend: Inactive worlds are automatically suspended to save resources
- Tab System: Unified tab management for web and 3D content
- ThreeJSWorldTab: Enhanced tab implementation for 3D world management
- ThreeJSTab Component: React component for Three.js 3D interface
- World Instance Manager: Manages multiple Three.js world instances
- Deep Search Interface: Advanced UI for autonomous web research
- AI Research Engine: Python backend for autonomous web browsing and analysis
- Model Integration: Universal adapter for all LLM providers and local models
- Three.js Bridge: WebGL-based 3D rendering and scene management
- Search Orchestrator: Coordinates multi-tab deep search operations
// Enhanced 3D world tab with Three.js integration
export class ThreeJSWorldTab extends BaseTab {
// Three.js world instance management
// Real-time rendering loop
// Scene manipulation via WebGL
// Resource cleanup and optimization
}// Autonomous web research with multi-tab coordination
export class DeepSearchEngine {
// Multi-page browsing automation
// Cross-reference analysis
// Context preservation across sessions
// AI-powered information synthesis
}// Universal LLM provider integration
export class ModelIntegrationService {
// gpt-oss model support
// Local model management (Ollama, vLLM)
// Cloud provider APIs
// Experimental 3D AI generation
}// electron/dev-config.js
module.exports = {
// Local model endpoints
ollama_base_url: "http://127.0.0.1:11434",
python_base_url: "http://127.0.0.1:8765",
// gpt-oss model settings
gpt_oss_120b_endpoint: "http://127.0.0.1:8000",
gpt_oss_20b_endpoint: "http://127.0.0.1:8001",
// Deep search configuration
max_concurrent_tabs: 10,
research_depth: 3,
enable_cross_reference: true,
// Experimental 3D AI
enable_text_to_3d: true,
three_js_renderer: "webgl2",
};// Maximum instances and resource limits
const config = {
maxInstances: 10,
maxMemoryPerInstance: 512, // MB
autoSuspendTimeout: 5 * 60 * 1000, // 5 minutes
enableMetaModelTracking: true,
enableScenePersistence: true,
};- Extend DeepSearchEngine: Add new autonomous browsing strategies
- Update Model Integration: Add support for new LLM providers
- Enhance Research UI: Add controls in research interface components
- Update Analysis Pipeline: Extend information synthesis capabilities
- Extend ThreeJSClient: Add new methods for 3D generation
- Update Types: Add new types in
types/api.ts - Enhance 3D UI: Add controls in
ThreeJSTab.tsx - Update AI Integration: Connect text-to-3D AI models
- Deep Search: Check browser console for research agent status
- 3D Worlds: Monitor Three.js performance and WebGL context
- AI Models: Check model endpoint connectivity and response times
- Multi-Tab Coordination: Use developer tools to monitor tab orchestration
- Enhanced 3D AI: Integration with advanced text-to-3D models
- Physics Simulation: Add realistic physics to Three.js worlds
- Export Capabilities: Save 3D scenes in standard formats (GLTF, OBJ)
- Collaborative Research: Multi-user deep search sessions
- Advanced 3D Generation: Full scene generation from natural language
- VR/AR Integration: Immersive 3D world exploration
- Knowledge Graphs: Visual representation of research findings
- AI-Generated Textures: Procedural material creation for 3D objects
- Fork the repository
- Create a feature branch
- Implement your changes
- Add tests for new functionality
- Submit a pull request
This project is licensed under the MIT License - see the LICENSE file for details.
For support and questions:
- Create an issue on GitHub
- Check the documentation
- Join our Discord community
OramBrowser - Where AI-powered research meets experimental 3D creation 🚀
Submission for: OpenAI Open Model Hackathon
Categories: Best Overall, Best Local Agent, Wildcard
Key Innovation: First browser to combine autonomous AI research with 3D world creation
Models Used: gpt-oss-120B, gpt-oss-20B, and comprehensive open-source model support
Demo Video: [Coming Soon]
Live Demo: [Coming Soon]
Built with ❤️ for the future of intelligent browsing and creative expression