Stagehand

Visit Website

Introduction

Stagehand is an open-source SDK for browser agents, built by Browserbase. The repository browserbase/stagehand is MIT-licensed, and the official site describes it this way: “Playwright was built for testing, Stagehand is built for agents.”

It offers familiar Playwright-style methods (goto, click, locator, screenshot) alongside natural-language calls such as act and extract, with self-healing actions. SDKs are available for TypeScript, Python and Go. At review the site showed version 4.

Features

1. Natural-language actions: act() performs an instruction such as clicking an element described in words; extract() pulls structured data.
2. Playwright-style API: goto, click, locator and screenshot are available for deterministic steps.
3. Self-healing actions: Actions can recover when the page changes, according to the site.
4. Token efficiency: A hybrid accessibility-tree trimming approach sends the model less page content; the site shows a comparison with plain Playwright.
5. Multi-language: TypeScript, Python and Go SDKs.
6. Agent features: WebMCP, clipboard support, batched commands, deep locators for nested iframes and OpenTelemetry support are listed.
7. Runs anywhere: Works with a local browser or Browserbase’s hosted browsers.

Pros

1. MIT licence: Permissive licence suits commercial test harnesses.
2. Mix of code and AI: Use deterministic code where you can and AI only where pages vary.
3. Popular project: Over 25,000 GitHub stars at review time.
4. Speed focus: The site publishes timing comparisons; verify on your own pages.
5. Agent-ready: Setup guides cover popular coding agents.

Cons

1. Not a test runner: There is no built-in test reporting; use it inside Vitest, Jest or Playwright Test.
2. Model costs: AI calls require an LLM provider and add per-run cost.
3. Variability: Natural-language actions can behave differently between runs.
4. Hosted browsers are paid: Browserbase’s cloud is a separate commercial service; see their site.
5. Fast-moving API: Major versions have changed APIs; pin versions.

Usage and Links (Example)

Install and run the quick start from the site:

npm install @browserbasehq/stagehand

const sh = await Stagehand.create();
const page = sh.context.activePage();
await page.goto("https://news.ycombinator.com");
await sh.act("click the first story");

Wrap these calls in your test runner and assert on extracted data. Check the docs for model configuration.

Resources:

• Stagehand website
• GitHub repository
• Docs

Pricing

Stagehand itself is free and open source under MIT. You pay your LLM provider for model calls, and Browserbase’s hosted browsers are billed separately; see browserbase.com for current plans.

Recommended for?

1. Engineers building AI-assisted end-to-end tests who want code-level control.
2. Teams testing pages that change often, where self-healing actions reduce maintenance.
3. Polyglot teams needing TypeScript, Python or Go.
4. Not a fit if you want a no-code product with dashboards and reports out of the box.

Conclusion

Stagehand is a well-maintained, permissively licensed SDK that blends Playwright-style code with natural-language browser actions.

It is a building block, not a full test platform. Use it inside your runner, keep critical assertions deterministic, and budget for model usage.