• 

  Oram Swity logo

Inspiration

The inspiration for OramBrowser came from a simple frustration: why do we still browse the web like it's 1995? When researching complex topics, I found myself manually opening dozens of tabs, struggling to synthesize information across sources, and losing context between sessions. Meanwhile, 3D creation tools remained expensive and isolated from the research process.

I envisioned a browser where AI agents could autonomously explore the web like a human researcher, opening multiple relevant pages, cross-referencing information, and maintaining context. Combined with experimental 3D creation capabilities, this would bridge the gap between information gathering and creative expression - perfect for showcasing the power of OpenAI's new gpt-oss reasoning models.

What it does

OramBrowser is the world's first AI-powered browser that combines autonomous web research with experimental 3D world creation:

🔍 Revolutionary "Open Tabs Tool":

• AI agents autonomously open and analyze multiple web pages simultaneously
• Cross-references information across sources for accuracy verification
• Builds comprehensive knowledge graphs from research sessions
• Maintains context across browsing sessions for continuous learning

🎮 Experimental 3D Creation:

• Real-time Three.js 3D world rendering directly in browser tabs
• Basic shape creation (boxes, spheres, cylinders) with plans for AI-generated objects
• Multiple isolated 3D environments with automatic resource management
• Future-ready architecture for advanced text-to-3D generation

🧠 Universal AI Integration:

• Supports all open-source models (Ollama)
• Specialized integration with gpt-oss-120B and gpt-oss-20B reasoning models
• Seamless cloud provider integration (OpenAI, Anthropic, Google)

How we built it

Architecture:

• Built with Electron, React, and TypeScript for cross-platform compatibility
• Implemented unified tab system managing both web research and 3D creation
• Created custom Three.js integration for browser-based 3D rendering
• Designed responsive UI with real-time research progress visualization

Model Integration:

• Universal adapter supporting all major LLM providers and local deployments
• Specialized gpt-oss model integration for advanced reasoning capabilities
• Experimental text-to-3D pipeline ready for future AI model improvements
• Local processing capabilities for complete offline operation

Technical Stack:

• Frontend: Electron + React + TypeScript + Three.js
• Models: gpt-oss (120B/20B), Ollama, local LLM deployments
• 3D: Three.js + WebGL for real-time rendering

Challenges we ran into

Multi-Tab Coordination Complexity: Managing autonomous AI agents opening and analyzing multiple web pages simultaneously proved incredibly complex. Ensuring proper context sharing, preventing resource conflicts, and maintaining research coherence across parallel browsing sessions required innovative coordination algorithms.

Three.js Performance Optimization: Running multiple 3D worlds in browser tabs while maintaining 60 FPS required careful resource management. We had to implement automatic instance suspension, memory optimization, and quality scaling to prevent browser crashes.

Universal Model Integration: Supporting every major LLM provider (local and cloud) while maintaining consistent behavior was challenging. Different models have varying response formats, capabilities, and rate limits that required extensive abstraction layers.

Context Preservation: Maintaining research context across browsing sessions, tab switches, and even app restarts required sophisticated state management. We built a persistence system that captures not just data, but the reasoning chains and connection patterns.

gpt-oss Integration: Being among the first to integrate OpenAI's new open-weight reasoning models meant working without established best practices. We had to experiment extensively to leverage their reasoning capabilities effectively for web research tasks.

Accomplishments that we're proud of

🚀 World's First AI-Autonomous Browser: We created the first browser where AI agents genuinely think and browse like humans, opening multiple tabs based on reasoning rather than simple search queries.

🧠 Advanced gpt-oss Showcasing: Successfully demonstrated novel applications of OpenAI's reasoning models beyond traditional chatbots, using them for complex multi-step web research and spatial understanding.

⚡ Seamless Integration: Achieved smooth integration between web research and 3D creation - users can research a topic and immediately start building related 3D concepts in the same interface.

🌐 Universal Model Support: Built truly universal AI integration supporting everything from local Ollama instances to cloud APIs, making advanced AI accessible regardless of infrastructure.

🎯 Real-World Problem Solving: Solved genuine productivity issues - our beta testers report 10x faster research workflows and improved information synthesis quality.

💡 Future-Proof Architecture: Created extensible systems ready for next-generation AI capabilities, especially as text-to-3D and reasoning models continue improving.

What we learned

AI Agent Coordination is Complex: Building truly autonomous AI agents that can coordinate multiple tasks (browsing, analysis, synthesis) taught us that the future of AI isn't just about better models - it's about better orchestration.

Context is Everything: The difference between useful and amazing AI applications lies in context preservation. Our cross-session context system became the most valuable feature for power users.

Local Models are Ready: Working extensively with local models showed us that privacy-focused, offline AI is not just possible but often preferable for sensitive research tasks.

Three.js + AI = Untapped Potential: The combination of browser-based 3D rendering with AI opens incredible possibilities we're only beginning to explore. The experimental features hint at a future where thought becomes immediate visual reality.

Open Source Models Excel: Through gpt-oss integration, we learned that open-weight models can match or exceed proprietary alternatives when properly fine-tuned for specific tasks.

What's next for OramBrowser

🎨 Advanced 3D AI Generation:

• Integration with cutting-edge text-to-3D models (Shap-E, Point-E successors)
• Full scene generation from natural language descriptions
• AI-powered texture and material creation
• Physics simulation and realistic object interactions

🧠 Enhanced Research Capabilities:

• Visual knowledge graph generation from research sessions
• Multi-language research support with real-time translation
• Collaborative research rooms for team investigations
• Academic paper analysis and citation networks

🌍 Platform Expansion:

• Mobile app with cloud synchronization
• Browser extension for traditional browsers
• VR/AR integration for immersive 3D world exploration
• API for developers to build custom research tools

⚡ Performance & Scale:

• Distributed AI processing for faster research
• Advanced caching for instant context retrieval
• Real-time collaboration features
• Enterprise deployment options

🚀 Community & Ecosystem:

• Open marketplace for custom research agents
• Community-contributed 3D templates and tools
• Educational partnerships for academic research
• Developer SDK for building on OramBrowser

The future of browsing isn't just about consuming information - it's about thinking, creating, and exploring in ways we never imagined possible. OramBrowser is just the beginning! 🚀

Built With

• electron
• next.js