TopoClaw: An open-source agent OS for cross-device, multi-user collaboration
New OS treats agents as 'Digital Twins' with topology-aware permissions across devices and teams.
TopoClaw is a new open-source Agent Operating System (Agent OS) designed to manage autonomous AI agents across distributed, collaborative, and permission-sensitive workflows. Unlike existing agent-centric OS designs that treat agents as single-host runtimes for internal reasoning and tool use, TopoClaw takes a human-centric and topology-aware approach. It models the user's ecosystem as two coupled structures: a physical device topology (heterogeneous surfaces like phones, laptops, smart home devices) and a social relationship topology (shared spaces, teams, delegated roles). This allows agents to operate seamlessly across devices and users while maintaining context, accountability, and security.
The system introduces three core innovations. First, cross-device action placement decouples intent from actuation, routing distributed actions across a device cluster based on hardware affordances and user context. Second, cross-user identity attribution treats agents as socially situated 'Digital Twins' that coordinate in multi-user spaces while preserving provenance, role-aware permissions, and human accountability. Third, cross-context authority governance pairs broad capability with distributed, context-aware policy enforcement across physical and social trust boundaries. This paper, from a team of 10 researchers and published on arXiv (2026), serves as an engineering-oriented reference architecture covering runtime, cross-device execution, collaboration mechanisms, security model, and deployment outlook.
- Models user ecosystem as two topologies: physical device topology and social relationship topology for context-aware agent behavior.
- Introduces cross-device action placement that decouples intent from actuation, routing actions based on hardware affordances and user context.
- Uses cross-user identity attribution with 'Digital Twins' for multi-user coordination, preserving provenance and role-based permissions.
Why It Matters
Brings accountability and human control to multi-device, multi-user agent workflows, enabling practical enterprise and home automation.