Mobile Agent

Benchmarks, environments, safety tests, and training recipes for phone-use agents operating real mobile workflows.

Phone-use agents and mobile workflow execution
Mobile Agent Phone-use agents GUI agents Privacy and safety Agent training
PhoneHarness mobile agent overview

Mobile Agent builds the infrastructure, benchmarks, and training recipes for phone-use agents that can operate mobile workflows reliably, privately, and safely. The research line moves beyond screenshot QA toward executable tasks, auditable traces, mixed action spaces, simulated phone environments, and real-app reinforcement learning.

Background and Motivation

Phones are a natural agent arena

Daily work already happens across mobile apps, permissions, notifications, files, accounts, and dynamic screens, making phones a realistic testbed for general-purpose agents.

GUI-only evaluation is too narrow

Real workflows often require a mixture of GUI interaction, command-line execution, tool calls, state inspection, and verification beyond simple click trajectories.

Success must be safe and private

An agent that completes a task while leaking data, overusing permissions, or taking unsafe actions is not deployment-ready.

Research Storyline

Privacy
Measure whether phone agents respect user data

MyPhoneBench reframes phone-agent readiness around permissioned access, minimal disclosure, and user-controlled memory.

Safety
Separate safe behavior from incapability

PhoneSafety evaluates safety-critical moments and distinguishes safe actions, unsafe actions, and failures to do anything useful.

World
Scale controllable phone environments

PhoneWorld turns real GUI trajectories and screenshots into reusable environments with executable tasks, state transitions, and verifiers.

Harness
Combine GUI, CLI, and tool actions

PhoneHarness gives phone agents a mixed-action execution framework with auditable traces and task-level verification.

Train
Train open models for real phone use

PhoneBuddy combines real-app environments and mock environments so open models can learn robust agentic phone-use behavior.

Representative Work

Privacy
Do Phone-Use Agents Respect Your Privacy?

Introduces MyPhoneBench for evaluating privacy risks in phone-use agents across realistic mobile tasks.

Paper
Safety
Safe, or Simply Incapable?

Rethinks safety evaluation by separating genuinely safe choices from unhelpful or failed actions.

Paper
Env
PhoneWorld: Scaling Phone-Use Agent Environments

Builds scalable phone-use environments from real GUI traces, controllable screens, executable tasks, and verifiers.

Paper
Harness
PhoneHarness: Harnessing Phone-Use Agents through Mixed GUI, CLI, and Tool Actions

Expands agent execution beyond GUI-only actions and evaluates whether mobile workflows are truly completed.

Paper
Training
PhoneBuddy: Training Open Models for Agentic Phone Use

Uses real-app and mock phone environments to train open models for stronger agentic phone-use performance.

Paper

Technical Layers

Environment construction

Reusable mobile environments make real app workflows controllable, replayable, and measurable.

Mixed action execution

Agents can combine GUI operations, CLI commands, and tool calls while leaving traces that humans and verifiers can audit.

Task-level verification

Benchmarks judge whether the external phone state changed correctly, not just whether the model produced plausible intermediate actions.

Safety and privacy constraints

Evaluation tracks permission use, data disclosure, memory control, unsafe operations, and refusal behavior in mobile settings.

Display Figures

Resource Map

PhoneHarness

Project site for mixed-action phone-use agent harnessing and benchmark resources.

Project site
PhoneHarness code

Open-source implementation for the phone-use harness and evaluation workflow.

Repository
PhoneHarness Bench

Dataset release for evaluating phone agents through mixed GUI, CLI, and tool-action traces.

Dataset