Touchy

Product Introduction

1. Definition: Touchy is a multimodal AI assistant application for iOS, designed to operate as a context-aware, proactive agent. It technically falls into the categories of personal AI, ambient computing, and agentic AI.
2. Core Value Proposition: Touchy exists to reduce smartphone dependency and digital distraction by providing a quiet, efficient AI that handles tasks autonomously. Its primary value is enabling users to "stay in the moment" by minimizing interaction time—ask a question or give a command, then put the phone away while the assistant works in the background.

Main Features

1. Multimodal Context Awareness: Touchy is engineered to perceive and act within the user's physical and digital environment. It utilizes on-device audio and video processing to understand real-world context, and employs secure API integrations to interact with other apps and services. This allows it to "see what you see and hear what you hear" to provide relevant assistance.
2. Proactive, Minimal-Interaction Interface: The core UX philosophy is asymmetric interaction—minimal output for the user, maximal information intake for the AI. The system uses theory-of-mind modeling to detect when a user is uncertain or needs help, aiming to provide a single, helpful signal before working autonomously, thereby reducing screen time and notification fatigue.
3. Research-Driven Agentic Architecture: Unlike retrieval-based chatbots, Touchy is built as an autonomous agent with a focus on safety and alignment. It employs a "harness" that controls its actions to prevent reward hacking or unauthorized operations. The architecture is designed for continual learning, allowing the assistant to adapt to user preferences and dialects over time through weight adjustments, context management, and recursive agentic structures.

Problems Solved

1. Pain Point: It addresses the "clunky and unhelpful" nature of current consumer AI assistants, which are often glorified search tools requiring constant manual input and screen engagement, thus pulling users out of the physical world.
2. Target Audience: The primary personas are professionals and individuals seeking digital wellness, who are overwhelmed by app fragmentation (needing 10+ apps daily) and desire seamless, ambient assistance. Secondary audiences include early adopters interested in cutting-edge agentic AI and long-horizon multimodal reasoning applications.
3. Use Cases: Essential scenarios include: getting immediate, context-aware answers without unlocking your phone; having the assistant monitor a long-term physical task (like assembling furniture or cooking) via video and provide step-by-step guidance; automating multi-app workflows (e.g., checking calendar, weather, and traffic to suggest departure time) with a single voice command.

Unique Advantages

1. Differentiation: Unlike Siri, Google Assistant, or even ChatGPT, which are primarily reactive or chat-based, Touchy is built from the ground up as a proactive, autonomous agent focused on completing tasks with minimal user interaction. It contrasts with "addiction-forming" apps by prioritizing disengagement after providing value.
2. Key Innovation: Its unique bet is the deep integration of foundational AI research (continual learning, long-horizon reasoning, theory of mind, AI alignment) directly into a consumer product UX. The hybrid privacy approach, combining efficient on-device models for sensory processing with secure cloud delegation for complex reasoning, is also a key technical innovation in the personal AI space.

Frequently Asked Questions (FAQ)

1. How does Touchy protect my privacy while processing audio and video? Touchy employs a hybrid AI architecture where initial sensory processing (raw audio/video) is handled by efficient on-device models to filter and anonymize data before any sensitive information is potentially delegated to more powerful cloud models for complex reasoning, keeping user data control paramount.
2. What makes Touchy an "AI agent" and not just another chatbot? Unlike chatbots that primarily retrieve and generate text, Touchy is an autonomous AI agent capable of taking actions. It operates through a controlled "harness" that allows it to interact with other apps and services on your behalf to complete multi-step tasks, based on long-horizon reasoning about your goals.
3. Can Touchy learn my personal preferences and habits over time? Yes, a core research focus of Touchy is continual learning. The system is designed to adapt to your unique lexicon, preferences for detail, and routine over time through various methods, including updating its internal context and learning from observed interactions to become more personalized and helpful.
4. Is Touchy available for Android or as a web app? Currently, Touchy is only available as a native iOS application on the Apple App Store. The focus on deep system integration and ambient, on-device processing aligns closely with the iOS ecosystem, with no announced timeline for Android or web versions.
5. How does Touchy's "put your phone down" philosophy work in practice? The assistant is designed for asymmetric interaction. You can activate it with a quick voice query or gesture. It then uses its multimodal sensors and agentic capabilities to gather necessary information and execute tasks, delivering only a concise, actionable result (e.g., a haptic buzz or brief sound) to minimize the need for you to look at the screen.