Researchers have introduced PalmClaw, an open-source agent framework designed to run natively on mobile devices rather than relying on external servers or GUI automation. The system manages sessions, memory, skills, and tools directly on the phone, exposing device capabilities as explicit tools with structured results.
- PalmClaw allows agents to access mobile hardware and applications directly through defined execution boundaries, avoiding the limitations of graphical user interface actions like tapping and swiping.
- The framework provides a clear separation between agent logic and device control, enabling more reliable task automation on smartphones.
- Experimental results demonstrate an 11.5% relative improvement in task success rates and a 94.9% reduction in completion time compared to the strongest baseline.
This approach addresses the challenges of existing mobile agents by providing direct access to device capabilities while maintaining explicit control over each action, making it suitable for users who require efficient and secure on-device automation.