Practical overview
When it fits
Use Mobilegym when mobile-agent research needs controllable simulated apps, inspectable environment state and repeatable task judgments, especially for experiments requiring parallel runs.
Who it is for
- Researchers benchmarking agents that operate Android-style graphical interfaces.
- Teams studying reproducible task evaluation and unintended agent side effects.
- Researchers conducting parallel mobile-agent rollouts or online reinforcement-learning experiments.
Documented uses
- Benchmark mobile GUI agents on simulated app tasks.
- Check whether an agent achieved a task by evaluating structured environment state.
- Compare environment state to identify unintended side effects.
- Run parallel simulations for evaluation or research rollouts.
- Use the environment in online reinforcement-learning experiments.
Requirements & limits
The repository is public and identifies its code license as Apache-2.0. The companion dataset is listed as CC BY-NC 4.0. The README describes a live browser demo and local setup; it does not establish account or pricing terms.
- The repository documents 28 simulated apps and 416 task templates.
- Task outcomes have programmatic judges that read structured environment state.
- The README reports a distinction between local serving modes: development/manual exploration, smaller single-agent evaluation, and a gateway mode for heavier benchmark or RL workloads.
- Setup spans a Node.js frontend, Python benchmark dependencies, Playwright Chromium, and a separate companion dataset.
- Simulator keys are optional for the canonical benchmark, while some additional capabilities use keys.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Install Mobilegym and start the local simulator
Prepare the documented frontend and Python benchmark dependencies, then start the simulator for manual exploration.
Before you start
- Node.js 22 or later
- Python 3.11 or later
- Chromium installed through Playwright
- The companion dataset described in the README
Steps
Clone the Mobilegym repository and change into its directory.
Run the command below for the frontend.
npm installInstall Python dependencies with the command below, then run
playwright install chromium.pip install -r bench_env/requirements.txtDownload and extract the companion dataset using the release URL and commands in the README.
Run the command below to serve the simulator at
http://localhost:3000.npm run dev
Expected outcome
A locally served simulator for manual exploration and development, with the benchmark runtime dependencies and dataset prepared.
Guide limitations
- The README identifies the dataset as CC BY-NC 4.0; review its license and disclaimer.
- This recipe covers setup and local exploration, not running a benchmark or training an agent.
- Simulator keys are optional for the canonical benchmark; additional key-dependent capabilities are described in the repository documentation.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 13 October 2026
- Original website title
- Purewhiter/mobilegym · Official homepage title · retained in its source language

