UI-TARS Desktop
ByteDance's desktop agent: the UI-TARS vision model moves your real mouse — and only that model family can drive it
What is UI-TARS Desktop?
Type an instruction and the app screenshots your screen, feeds it to the UI-TARS vision-language model, and issues the clicks, keystrokes and scrolling to carry it out — across the whole desktop or a browser, which is what separates it from browser-only agents. The model is the catch: you bring an endpoint serving the UI-TARS or Doubao family — self-hosting the open 7B weights via vLLM is the documented local route — and frontier models like GPT or Claude cannot drive the desktop app. Weigh the project's pulse too: desktop binaries have not shipped since August 2025, the hosted remote operators were discontinued, and 2026 activity slowed to a trickle as the repository pivoted toward its Agent TARS stack.
What can you do with UI-TARS Desktop?
- The desktop, not just the browser — A computer operator drives the whole screen and a browser operator drives Chrome, Edge or Firefox — file managers, office software and native dialogs are all reachable. It moves the real cursor with no sandbox in between, and macOS requires Accessibility and Screen Recording permissions before it can act.
- You bring the model endpoint — Any OpenAI-compatible endpoint serving the family works: the open UI-TARS-1.5-7B weights on your own GPU via vLLM or a Hugging Face inference endpoint, or ByteDance's VolcEngine cloud serving its Doubao vision models.
- Agent TARS, the general-purpose sibling in the same repo — An npm CLI with a local web UI turns the same stack into a broader agent for search, browsing and file work, with three browser-control modes — a DOM mode that works with non-vision models, visual grounding from screenshots, and a hybrid of the two.
- An SDK for building your own operators — The UI-TARS SDK ships operators for desktop input, browsers and Android over adb, so the screenshot-to-action loop can be embedded in your own automation rather than used only through the app.
Before you choose UI-TARS Desktop
- Only UI-TARS and Doubao endpoints are documented for the desktop app; Ollama support is the tracker's most-reacted open request and GPT or Claude cannot substitute.Reported in#43
- Windows and macOS only — Linux has been an open request since January 2025.Reported in#9
- Distribution has stalled: the last desktop binary is from August 2025, nothing has shipped since November 2025, and the hosted remote operators were switched off — check the pulse before committing.
Star history
19 Aug to 28 Aug · +94
Frequently asked questions
Is UI-TARS Desktop free for commercial use?
UI-TARS Desktop is released under the Apache-2.0 licence — OSI-approved open source, which permits commercial use.
How can UI-TARS Desktop be deployed?
UI-TARS Desktop is available as Runs locally.
Documentation
Reproduced from the bytedance/UI-TARS-desktop README, published under Apache-2.0. Read the original ↗
Introduction
English | 简体中文
TARS* is a Multimodal AI Agent stack, currently shipping two projects: Agent TARS and UI-TARS-desktop:
Table of Contents
News
- [2025-11-05] 🎉 We’re excited to announce the release of Agent TARS CLI v0.3.0! This version brings streaming support for multiple tools (shell commands, multi-file structured display), runtime settings with timing statistics for tool calls and deep thinking, Event Stream Viewer for data flow tracking and debugging. Additionally, it features exclusive support for AIO agent Sandbox as isolated all-in-one tools execution environment.
- [2025-06-25] We released an Agent TARS Beta and Agent TARS CLI - Introducing Agent TARS Beta, a multimodal AI agent that aims to explore a work form that is closer to human-like task completion through rich multimodal capabilities (such as GUI Agent, Vision) and seamless integration with various real-world tools.
- [2025-06-12] - 🎁 We are thrilled to announce the release of UI-TARS Desktop v0.2.0! This update introduces two powerful new features: Remote Computer Operator and Remote Browser Operator—both completely free. No configuration required: simply click to remotely control any computer or browser, and experience a new level of convenience and intelligence.
- [2025-04-17] - 🎉 We’re thrilled to announce the release of new UI-TARS Desktop application v0.1.0, featuring a redesigned Agent UI. The application enhances the computer using experience, introduces new browser operation features, and supports the advanced UI-TARS-1.5 model for improved performance and precise control.
- [2025-02-20] - 📦 Introduced UI TARS SDK, is a powerful cross-platform toolkit for building GUI automation agents.
- [2025-01-23] - 🚀 We updated the Cloud Deployment section in the 中文版: GUI模型部署教程 with new information related to the ModelScope platform. You can now use the ModelScope platform for deployment.
Agent TARS
Agent TARS is a general multimodal AI Agent stack, it brings the power of GUI Agent and Vision into your terminal, computer, browser and product. It primarily ships with a CLI and Web UI for usage. It aims to provide a workflow that is closer to human-like task completion through cutting-edge multimodal LLMs and seamless integration with various real-world MCP tools.
Showcase
Please help me book the earliest flight from San Jose to New York on September 1st and the last return flight on September 6th on Pricelinehttps://github.com/user-attachments/assets/772b0eef-aef7-4ab9-8cb0-9611820539d8
For more use cases, please check out #842.
Core Features
- 🖱️ One-Click Out-of-the-box CLI - Supports both headful Web UI and headless server execution.
- 🌐 Hybrid Browser Agent - Control browsers using GUI Agent, DOM, or a hybrid strategy.
- 🔄 Event Stream - Protocol-driven Event Stream drives Context Engineering and Agent UI.
- 🧰 MCP Integration - The kernel is built on MCP and also supports mounting MCP Servers to connect to real-world tools.
Quick Start
# Launch with `npx`.
npx @agent-tars/cli@latest
# Install globally, required Node.js >= 22
npm install @agent-tars/cli@latest -g
# Run with your preferred model provider
agent-tars --provider volcengine --model doubao-1-5-thinking-vision-pro-250428 --apiKey your-api-key
agent-tars --provider anthropic --model claude-3-7-sonnet-latest --apiKey your-api-keyVisit the comprehensive Quick Start guide for detailed setup instructions.
Documentation
🌟 Explore Agent TARS Universe 🌟
UI-TARS Desktop
UI-TARS Desktop is a native GUI agent for your local computer, driven by UI-TARS and Seed-1.5-VL/1.6 series models.
Showcase
| Instruction | Local Operator | Remote Operator |
|---|---|---|
| Please help me open the autosave feature of VS Code and delay AutoSave operations for 500 milliseconds in the VS Code setting. | ||
| Could you help me check the latest open issue of the UI-TARS-Desktop project on GitHub? |
Features
- 🤖 Natural language control powered by Vision-Language Model
- 🖥️ Screenshot and visual recognition support
- 🎯 Precise mouse and keyboard control
- 💻 Cross-platform support (Windows/MacOS/Browser)
- 🔄 Real-time feedback and status display
- 🔐 Private and secure - fully local processing
Quick Start
See Quick Start
Other options in Computer & Browser Use
Browser UseComputer UseExtracts the interactive elements of a page and hands the model a structured list, instead of asking it to guess click coordinates from a screenshot. That makes it markedly more reliable for form flows and scraping. It degrades on canvas-heavy applications and on sites with aggressive bot protection, where there is no clean DOM to read.
openinterpreterCoding AgentsType `i` in a project directory and it reads your files, edits them and runs commands inside a sandbox you configure. What sets it apart is `/harness`: it swaps the whole model-facing surface — system prompt, tool schema, message conversion — for the one a given vendor tuned its model on, which is how it aims to get usable work out of inexpensive models like DeepSeek or Kimi. The trade-off is what you are actually adopting: this is a fork of OpenAI's Codex CLI, rewritten in Rust and still on the 0.0.x releases it started in June 2026, sharing little more than a name and a repository with the Python Open Interpreter most people have heard of.
playwright-mcpComputer UsePlaywright MCP puts a real Chrome, Firefox, WebKit or Edge behind an MCP server and returns each page as an accessibility snapshot: a text outline of roles, names and reference ids that the agent clicks and types into, with no vision model and no coordinate guessing. Past plain navigation it can mock network requests, read and write cookies and web storage, record a trace or a video, and emit Playwright locators and assertions, which makes it a practical way to turn a session you drove by hand into a test you can check in. The cost is context: Microsoft's own README points coding agents at the Playwright CLI with skills instead, because the tool definitions and the snapshots compete for the same window as your codebase. It earns its keep in long-running loops, where holding one browser open across many turns is worth paying that.