AI AGENTS
Webwright
by Microsoft
Overview
Terminal-native web-agent harness where models use Playwright code and bash commands for long-horizon web tasks.
Details
Webwright is a Microsoft project described as a simple SWE-style browser agent framework for long-horizon web tasks. Microsoft Research describes it as a minimal terminal-based web-agent harness where models write Playwright code and bash commands instead of predicting one browser click at a time. The official project page says Webwright gives models a terminal, local workspace, and code-writing freedom to launch, inspect, and discard browser sessions, producing reusable programs.
When to Use
Use when evaluating terminal-native approaches to web agents that operate through code and shell commands rather than direct click prediction. Use when researching browser-agent frameworks for long-horizon web tasks and reusable Playwright-based programs.
Getting Started
- Open the official project page at https://microsoft.github.io/Webwright/ to understand the terminal-native web-agent approach.
- Review the GitHub repository at https://github.com/microsoft/Webwright for source code and project materials.
- Read the Microsoft Research article for context on the design: https://www.microsoft.com/en-us/research/articles/webwright-a-terminal-is-all-you-need-for-web-agents/.
Key Features
- •Terminal-based web-agent harness.
- •Models write Playwright code and bash commands instead of predicting one browser click at a time.
- •Provides a terminal and local workspace for browser-agent workflows.
- •Supports launching
- •inspecting
- •and discarding browser sessions.
- •Designed to produce reusable programs for web tasks.
Capabilities
- •browser-agent-framework
- •terminal-native-agent-harness
- •playwright-code-generation
- •long-horizon-web-tasks
Last updated Jun 4, 2026