Touchy is developing an AI assistant that inhabits the user's physical and digital context, using multimodal inputs like video and audio to understand and act on real-world situations. The mobile app aims to replace the textbox interface with a system that can see, hear, and interact with the user's environment, allowing for continuous learning and long-horizon reasoning across tasks like assembling furniture or cooking. The company focuses on building agent harnesses that connect AI to the world while improving alignment with user intentions.
Funding
Funding not disclosed
Founders
Product
Problem
Consumer AI assistants are built around retrieval and text-based interfaces, forcing users to open multiple apps daily to manage tasks. These tools lack the ability to perceive and interact with the physical world, making them unhelpful for real-world activities. There is also a persistent challenge of AI systems failing to reliably understand and preserve user intentions when acting autonomously.
Solution
Touchy provides an AI assistant that connects to the user's physical and digital context through a mobile application. The system captures what the user sees and hears, enabling it to offer assistance across real-world scenarios like assembling furniture, cooking, or navigating unfamiliar environments. Touchy's underlying architecture incorporates long-horizon multimodal reasoning to process continuous video and audio inputs, allowing it to provide helpful information over extended periods. The platform emphasizes continual learning, adapting to a user's preferences and information over time. By combining agent harnesses that integrate with available tools and systems, the product facilitates autonomous action that aligns with user intentions.
Target Audience
Primary users are consumers seeking a practical AI assistant that can help with everyday physical tasks, hobbies, and real-world problem-solving. The product also targets individuals who prefer an alternative to traditional text-based AI interfaces.
Features
- Mobile application with multimodal input capabilities for video and audio processing
- Continual learning system that adapts to user preferences through weight/architecture changes, context modifications, and agent self-modification
- Long-horizon reasoning to maintain context and provide assistance across extended real-world tasks and processes
- Agent harnesses designed to connect seamlessly with physical tools, digital systems, and services
- Privacy-focused data handling with no sensitive information processing and transparent user controls