Idea
An OS-level harness, forked from Android, built to treat AI agents as first-class citizens instead of app-guests operating through interfaces made for human fingers.
Research Paper
Core Innovation
This paper introduces AOHP, an OS-level agent harness built on Android that treats AI agents as first-class OS actors. It innovates by enabling personalized service composition, efficient agent interfaces, and secure information flow, improving task completion and reducing execution costs compared to conventional systems.
Why It Matters
Phone operating systems assume one human tapping one app at a time. That assumption creates real overhead and safety gaps once an agent is doing the tapping. Agents get structured app access instead of screenshots, background execution that doesn't hijack the screen, and sensitive data swapped for placeholder tokens before an agent ever sees it. It beats stock Android meaningfully on task completion and cost, and the code is public. What it doesn't cover is the actual product: an enterprise data-security layer for agents on managed devices, a lighter SDK for mobile agent builders, or a narrow hardware play like kiosks and field devices. One hard constraint worth flagging: this can't ship as a normal app download. It needs OEM cooperation or managed-device deployment to reach real users.
Market Size (TAM)
$20–50B TAM for AI-enhanced operating systems; $2–10B SAM from mobile OS and enterprise software sectors. Driven by rising AI agent adoption and demand for secure, efficient AI workflows.
Potential Customers & Pain Points
- Mobile OS developers – Need native AI agent support
- Enterprise software providers – Need efficient multi-application AI workflows
- Security-conscious organizations – Need secure AI agent execution environments
Business Model
Open-source platform with enterprise licensing for advanced features, customization, and support services.
Competitive Landscape
- Microsoft Copilot
- Apple SiriKit
- Google Assistant SDK
Implementation Challenges
- Integration complexity with existing OS ecosystems
- User adoption of agent-native OS features
- Ensuring robust security and privacy compliance
Validation Strategy
- Pilot deployments with mobile OS developers
- Performance benchmarking against conventional OS agent workflows
- Security audits and compliance testing
Research Paper Overview
AOHP: An Open-Source OS-Level Agent Harness for Personalized, Efficient and Secure Interaction
Summary
AI agents can autonomously manage tasks across applications but current OSes lack native support, causing inefficiencies and security risks. AOHP integrates agents as first-class OS actors on Android, improving task completion, reducing execution costs, and enhancing security compliance.