Breaking

ByteDance Ships Agent TARS CLI and Web UI for Multimodal GUI Automation

September 16, 2026 at 09:34 EDT

AI Agents
Open Source
Foundation Models

ByteDance's TARS project has shipped Agent TARS as an open-source multimodal agent stack that brings GUI control and vision into terminal, browser and product workflows through a CLI and Web UI. The stack is built around human-like task completion, combining visual grounding, browser control and Model Context Protocol tooling rather than limiting agents to chat or code-only interfaces.

Sept. 16, 2026 / Agent TARS

GUI agents move into the terminal

Agent TARS packages visual computer control, browser automation and MCP tool access behind a CLI and Web UI, aiming to make computer-use agents usable in developer workflows.

Latest release

v0.3.0

Repository stars

39,002

Forks

3,945

OSWorld benchmark jump in the model line

24.6UI-TARS

47.5UI-TARS-2

The later technical report nearly doubles the earlier OSWorld score, while the product stack still has to prove reliability on messy real interfaces.

Why developers care

Web research, browser automation, data analysis, UI recreation and report generation can run through one multimodal agent workflow.

What remains hard

Long context, specialized knowledge and visual grounding errors are still named as difficult cases for real use.

Agent workflow

CLI or Web UI→Hybrid browser control→MCP tools→Workspace output

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.