Do I need to install a browser to use Browserbase?

No. The browser runs in the cloud. start creates or reuses a Browserbase session and the agent drives it from there, so a serverless function or CI runner can automate the web without local Chromium. Playwright, the top pick, is the option when you do want to run the browser locally.

How does the agent control the page without writing selectors?

Through Stagehand. The act tool performs a natural-language action on the page and observe lists the actionable elements first, so the agent describes intent instead of hand-coding selectors. extract then pulls structured data from the page given an instruction.

Browserbase for browser automation

Pick 2 of 3 for browser automationOfficialBrowserbase3,364

Browserbase runs the browser in the cloud and drives it with Stagehand's natural-language actions, so an agent can automate the web from a serverless function or a CI runner with no local Chromium to install or sandbox. For browser automation it is the second of three picks, the right choice when you want the automation to run somewhere other than the agent's own machine.

It ranks second because the leading pick is the more common default for local, free, self-driven automation, but Browserbase wins a real lane: removing the infrastructure entirely. The browser lives in the cloud, you pay per session, and the agent describes what to do in plain language rather than scripting selectors.

How Browserbase fits

A session starts with start, which creates or reuses a Browserbase cloud session, and ends with end. navigate points the cloud browser at a URL. The interesting tools are the Stagehand pair: act performs a natural-language action on the page ("click the login button," "fill the search box"), and observe returns the actionable elements and candidate actions when the agent needs to see what is possible before acting. extract pulls structured data from the current page given an instruction. That natural-language layer means the agent can automate without hand-written selectors.

Playwright is the first pick because it is the free, local, self-driven default: you run the browser yourself with fine-grained scripted control and no per-session cost. Exa is the simpler tool when the job is really content extraction from pages rather than driving a flow, since search-and-content retrieval avoids a browser altogether. Choose Browserbase when you need cloud browsers with no local infrastructure and want Stagehand's natural-language actions; choose Playwright when local control and zero session cost matter more.

Tools you would use

Tool	What it does
start	Creates or reuses a Browserbase cloud session for the agent to drive.
end	Closes the current Browserbase session.
navigate	Navigates the cloud browser to a URL.
act	Performs a natural-language action on the page via Stagehand.
observe	Observes the actionable elements on the page and returns candidate actions.
extract	Extracts structured data from the current page given an instruction.

Full Browserbase setup and config →

FAQ

Do I need to install a browser to use Browserbase?: No. The browser runs in the cloud. start creates or reuses a Browserbase session and the agent drives it from there, so a serverless function or CI runner can automate the web without local Chromium. Playwright, the top pick, is the option when you do want to run the browser locally.
How does the agent control the page without writing selectors?: Through Stagehand. The act tool performs a natural-language action on the page and observe lists the actionable elements first, so the agent describes intent instead of hand-coding selectors. extract then pulls structured data from the page given an instruction.