Skip to the content
Know your tech before you buy it. Makers: get listed from $49/mo

Computer Use OOTB

github.com

  • Free plan
  • Windows
  • macOS
  • Web
  • Self-hosted
8.1 Club score Visit github.com

What Computer Use OOTB is

Computer Use OOTB is an open-source framework for automating desktop tasks through a graphical interface. It provides a Gradio interface and works with API-based or locally running models; Docker is not required. Supported model options include Claude 3.5 Sonnet, GPT-4o, Qwen models, ShowUI, and UI-TARS. Users can assign one model to a task or configure separate models for planning and computer control. Local inference can avoid API costs, though a task using Claude 3.5 Computer Use may cost a few dollars. The setup instructions call for Miniconda, Python 3.12 or newer, and dependency installation. ShowUI local runs require a CUDA-enabled NVIDIA GPU with at least 6 GB of GPU memory on Windows, or an M1-or-newer Apple Silicon Mac with at least 16 GB of unified RAM. A mobile browser can send commands to the computer running the framework without a mobile app. The project warns that models can take unintended or harmful actions and recommends monitoring them. It also cautions that anyone with a public Gradio link may be able to control the computer. The project is free and licensed under Apache-2.0.

Who it suits

It suits people seeking to automate desktop or browser tasks on Windows or macOS, using API-based or local models. Local ShowUI use requires the specified hardware, and users should be prepared to monitor model actions.

Good to know, and what to watch

  • Free and licensed under Apache-2.0.
  • Works with API-based and locally running models.
  • Mobile browsers can send commands without a mobile app.
  • Gradio interface does not require Docker.
  • Setup requires Miniconda and Python 3.12 or newer.
  • ShowUI local runs have specific hardware requirements.
  • A public Gradio link may let others control the computer.
  • Claude 3.5 Computer Use tasks may cost a few dollars.

Open-source software

Free

Open source

Getting started

  1. Open the project’s GitHub page
  2. Install Miniconda
  3. Use Python 3.12 or newer
  4. Install the dependencies
  5. Start the interface with `python app.py`
  6. Configure an API-based or local model

Limits to check

Local ShowUI runs require a CUDA-enabled NVIDIA GPU with at least 6 GB of memory on Windows, or an M1-or-newer Apple Silicon Mac with at least 16 GB unified RAM. A Claude 3.5 Computer Use task may cost a few dollars.

The Club's review

Overview

Computer Use OOTB is an open-source framework for automating desktop work through graphical interfaces. It is best suited to technically comfortable users who want to select cloud or local models and shape how planning and screen actions are divided.

This is a flexible foundation rather than a polished, unattended desktop appliance. Its combination of model choice, local inference and browser-based remote commands will appeal to developers and IT users; people wanting a managed, low-risk automation product should look elsewhere.

Key features

Model choice and roles

The framework supports Claude 3.5 Sonnet, GPT-4o, Qwen models, ShowUI and UI-TARS. You can use one model for a task or configure separate planner and actor models, assigning strategic decisions to one and low-level interface control to another. That is a useful degree of control for teams tuning an agent to different workloads, but model selection and configuration become part of the implementation effort.

Local models run on the computer and avoid API charges. Cloud use can still be expensive: a task using Claude 3.5 Computer Use may cost a few dollars. For local ShowUI runs, Windows requires a CUDA-capable NVIDIA GPU with at least 6 GB of GPU memory; macOS requires an Apple Silicon M1-or-newer Mac with at least 16 GB of unified RAM. Local inference therefore suits users with suitable hardware, while API models suit those willing to pay per task.

Interface and remote control

OOTB provides a Gradio interface and does not require Docker. Installation involves Miniconda, Python 3.12 or newer, dependency installation and launching python app.py. A device with internet access, including a phone browser, can send commands to the computer running the software without a mobile app. That is convenient for remote operation, but it increases the importance of access control.

Safety and deployment

The project supports hybrid deployment across Windows and macOS, with desktop control, browser control and screen understanding. Its README warns that models may take unintended or harmful actions and recommends monitoring them. It also warns that anyone with a generated public Gradio link may be able to control the computer; sharing that link is unsafe. This is not a suitable basis for unsupervised access to sensitive systems.

Pricing

PlanPriceWhat it includes
Open-source softwareFreeApache-2.0 license, API-based or locally running models, Windows and macOS

There is no subscription or seat charge. Free means the software itself costs nothing; API providers can still bill for usage, with Claude 3.5 Computer Use tasks potentially costing a few dollars each. Choosing local models removes API costs but shifts the burden to compatible hardware and setup.

Platforms

Computer Use OOTB is self-hosted on Windows or macOS and exposes a web interface through Gradio. It can be controlled from a mobile browser, but there is no requirement for a dedicated mobile application. The project is released under the Apache-2.0 license.

Who it's for

Choose OOTB if you are a developer, automation builder or IT operator who wants an open framework, configurable model roles and the option to run inference locally. It is particularly appropriate when avoiding recurring API charges matters and you have the hardware to support local models.

It is a poor fit for nontechnical users seeking a guided installer, for organisations needing unattended and tightly governed computer control, or for anyone without the required Python environment and local-model hardware.

Pros and cons

Pros

  • Flexible model architecture: supports several API and local models, with unified or separate planner and actor roles.
  • No software fee: Apache-2.0 open source avoids subscription costs and permits self-hosting.
  • Remote browser commands: a phone or other internet-connected device can operate the host without installing a mobile app.
  • Docker-free interface: Gradio launches directly after the documented Python setup.

Cons

  • Technical installation: Miniconda and Python 3.12 or newer are required before the interface can run.
  • Hardware barrier for local inference: ShowUI needs a qualifying GPU on Windows or substantial Apple Silicon memory on macOS.
  • Usage can cost money: API-based Claude computer-use tasks may cost a few dollars each.
  • Serious control risk: unintended model actions and exposed public Gradio links can put the controlled computer at risk.

Alternatives

Browse AI Computer Use Agents for a broader category view. Claude Code is the better choice if you want a paid service spanning web, desktop and mobile platforms rather than self-hosting this framework. Choose Verdesk for a Windows-focused freemium product with browser DOM mode. Gemini suits users who want a broader Google AI app with research, image tools and mobile access. Screenbox is another free, self-hosted option across Windows, macOS and Linux. Lapu AI offers a freemium desktop agent for Windows and macOS. agent-cu is a locally running, MIT-licensed alternative with zero per-action cost. ChatGPT for Excel is aimed at spreadsheet work in the browser. Scalaris Atlas is a free Windows, macOS and Linux option with zero egress by default.

Verdict

Computer Use OOTB is a strong choice for technically capable users who value open-source control, multiple model providers and local inference. Its main reason to choose it is flexibility without a software subscription. Look elsewhere if you need a turnkey product, predictable API costs or computer control that can safely run without close human supervision.

The verdict

Computer Use OOTB offers desktop automation with a choice of API-based and local models, plus remote commands from a mobile browser. Its setup and hardware requirements, and the risks associated with model actions and public links, are important to weigh.

Questions people ask

Is Computer Use OOTB free?

Yes. It is open source under the Apache-2.0 license.

Which operating systems are supported?

The README lists Windows and macOS.

Can models run locally?

Yes. Local models can run on the user’s device, with no API costs for local inference.

What models are supported?

The listed options include Claude 3.5 Sonnet, GPT-4o, Qwen models, ShowUI, and UI-TARS.

Can I control it from a phone?

A mobile browser can issue commands to the computer running OOTB, without a mobile app.

What are the setup requirements?

The instructions require Miniconda, Python 3.12 or newer, dependency installation, and launching `python app.py`.

What the maker publishes

Cost limit
The README says each task using Claude 3.5 Computer Use may cost a few dollars. source
Cost limit
The README says a Claude 3.5 Computer Use task may cost a few dollars; local models can run without API costs. source
Installation
The README instructs users to install Miniconda, use Python 3.12 or newer, install dependencies and start the interface with python app.py. source
Integrations
The documented API model keys include Anthropic, OpenAI and Qwen; local model setup also references PyTorch and UI-TARS deployment options. source
Interface
It provides a Gradio interface and does not require Docker. source
Interface
It provides a Gradio interface that runs locally and can be accessed through a browser. source
License
The project repository lists the Apache-2.0 license. source
License
The repository identifies its license as Apache-2.0. source
Local inference
The README says local models can run on the user's device with no API costs. source
Local model hardware
For ShowUI local runs, the README lists a CUDA-enabled NVIDIA GPU with at least 6 GB of GPU memory on Windows, or an M1-or-newer Apple Silicon Mac with at least 16 GB unified RAM. source
Local model requirements
For ShowUI local runs, the README specifies a CUDA-capable NVIDIA GPU with at least 6 GB GPU memory on Windows or Apple Silicon M1 or newer with at least 16 GB unified RAM on macOS. source
Maker affiliation
The project page identifies its authors with ShowLab at the National University of Singapore. source
Model options
The project supports API-based models and locally running models, including Claude 3.5 Sonnet, GPT-4o, Qwen models, ShowUI and UI-TARS. source
Model roles
Users can configure a unified model or separate planner and actor models for high-level planning and low-level control. source
Models
The README lists Claude 3.5 Sonnet, GPT-4o, Qwen models, ShowUI, and UI-TARS among supported GUI agent models. source
Operating systems
The README lists Windows and macOS as supported systems. source
Public link risk
The README warns that anyone with the generated public Gradio link may be able to control the computer and says not to share it. source
Purpose
Computer Use OOTB is an out-of-the-box GUI agent framework for automating desktop tasks with API-based and locally running models. source
Remote access
A user can issue commands from a mobile browser to control the computer running OOTB, without installing a mobile app. source
Remote access risk
The README warns that anyone with a public Gradio link may be able to control the computer and says not to share that link. source
Remote control
A device with internet access, including a mobile browser, can send commands to the computer running the software. source
Requirements
The setup instructions require Miniconda and Python 3.12 or newer. source
Security risk
The README warns that models can produce unintended or potentially harmful actions and recommends monitoring their actions. source
Security risk
The project warns that models may perform unintended or potentially harmful operations and recommends continuously monitoring their actions. source

Checked against the maker's pages on 9 Oct 2026.