An open-source Mac app that gives Claude Code a real browser, a live iOS Simulator and its own git worktree per chat.
Coding & DevelopmentFree
UI-TARS Desktop
by ByteDance
Released
Summary
ByteDance's open-source desktop app that lets the UI-TARS vision-language model operate your computer and browser with mouse and keyboard.
Screenshots
Description
What it is
UI-TARS Desktop is a free, open-source desktop application from ByteDance that lets a vision-language model operate your computer. You type an instruction such as "turn on autosave in VS Code settings", and the agent takes screenshots, recognises what is on screen and drives the mouse and keyboard to complete it. It is built on ByteDance's UI-TARS GUI-agent models (UI-TARS-1.5 and the Seed-1.5-VL/1.6 series) and lives in the UI-TARS-desktop repository alongside its sibling project, the Agent TARS CLI. The code is licensed under Apache 2.0.
Key features
- Computer Operator: controls your local machine directly through screenshots, visual recognition and precise mouse and keyboard actions.
- Browser Operator: automates a local Chrome, Edge or Firefox for web tasks such as navigating pages and filling in forms.
- Bring your own model: you point the app at a VLM endpoint, for example UI-TARS-1.5-7B deployed on Hugging Face Inference Endpoints or Doubao-1.5-UI-TARS on VolcEngine, by entering a base URL, API key and model name in Settings.
- Real-time feedback: a chat pane shows the task and the agent's status while it works, next to the screenshots it acts on.
- Cross-platform: macOS (Homebrew cask
ui-tarsavailable) and Windows builds; on macOS it needs Accessibility and Screen Recording permissions. - Local processing: the project describes the app as private and secure, with the agent running on your own machine.
Known limits from the docs: only single-monitor setups are supported, and the free hosted Remote Computer/Browser Operator introduced in v0.2.0 was discontinued on August 20, 2025.
Who it's for
Developers, researchers and tinkerers who want to try a GUI agent on their own desktop, compare GUI-agent models, or build on an open-source codebase rather than a hosted service. Because you have to deploy or subscribe to a model endpoint yourself, it suits people comfortable configuring API endpoints more than non-technical users.
Pricing
As listed in October 2026, the app is free and open source under the Apache 2.0 licence. You pay separately for whichever model endpoint you connect, such as a Hugging Face Inference Endpoint or VolcEngine's Doubao-1.5-UI-TARS API.
Reviews
No reviews yet — sign in to write the first one
Similar App Suggestions
See all UI-TARS Desktop alternatives
The open-source library that lets AI agents actually use websites — clicking, typing, and reading like a person.
Automation & WorkflowsFreemium
Open-source computer-use infrastructure: a desktop driver, cloud desktop fleets, local macOS VMs and a benchmark for agents that click.
Automation & WorkflowsFreemium
Free, open-source desktop browser that coding agents such as Claude Code, Codex and Cursor drive using your real logged-in accounts.
Automation & WorkflowsFree
A general autonomous AI agent that executes multi-step tasks end to end in its own cloud environment.
Automation & WorkflowsFreemium