Skip to content

Mobile-Agent-E

A mobile assistant that learns from past tasks

Visit official site

What Mobile-Agent-E is

Mobile-Agent-E is described as a mobile assistant built on large multimodal models that completes complex multi-step tasks through a hierarchical multi-agent framework with self-evolution.

What you can do with it

  • Run tasks through a hierarchy of manager, perceptor, operator, action reflector and recorder
  • Improve performance from tips and shortcuts in long-term memory
  • Handle long-horizon tasks that span multiple apps
  • Use detailed error feedback and automatic adjustment

Application scenarios

  • Complete a restaurant recommendation task with search, filtering and booking
  • Shop online by comparing products in several apps and completing a purchase
  • Arrange a travel itinerary with queries and bookings in multiple apps

Getting started

  • Visit the project page to review the functions and architecture
  • Download the open-source code and deploy it locally
  • Prepare task input such as a task description and an initial app screen
  • Run the agent and observe the high-level plan and actions it produces

Pros & cons

✓ What we like

  • Handles long-horizon tasks across multiple apps
  • Detailed error feedback and automatic adjustment are listed

! What to watch out for

  • The assistant is described as being at a research stage
  • Back-end work runs on supported large multimodal models

FAQ

What is the framework made of?

The material lists a manager, perceptor, operator, action reflector and recorder in a hierarchical multi-agent framework.

Which back-end models are named?

GPT-4o and Gemini are given as examples of supported large multimodal models.

Last reviewed: 2026-11-15

More AI office assistant tools

View all →

How we review