Browser Use

Browser Use is a Python framework for giving language-model agents a programmable web browser. It turns a natural-language task into browser actions such as opening pages, clicking, typing, uploading files and reading page content, while exposing the browser session and tool registry to application code. Developers can run the MIT-licensed library locally or self-host it, choose a supported model provider, add deterministic actions, validate final output with a Pydantic schema and reuse browser state when a workflow needs an authenticated session.

The Browser Use team also operates Browser Use Cloud, a separate hosted service built on managed browser infrastructure. Its Agent API accepts a goal and runs the web task, while its Browser API supplies a remote CDP endpoint for direct control from Playwright, Puppeteer, Selenium and other compatible clients. Cloud capabilities include profiles, proxies, live previews, run events, workspaces and human takeover. The open-source library is covered by the repository's MIT license; the hosted service has separate terms, usage charges and data policies.

Top features

• Agent and browser primitives: Compose model-driven task execution with configurable browser sessions, action limits, timeouts and fallback behavior.
• Bring your own model: Use Browser Use's model gateway or supported third-party and local model integrations rather than tying the agent loop to one provider.
• Custom actions and typed results: Register Python tools for application-specific operations and validate final results against Pydantic schemas.
• Local, existing or remote browsers: Run Chromium locally, connect to an existing profile, or attach to a remote browser over CDP.
• Cloud path when needed: Hosted agents and browsers add managed sessions, profiles, files, live observability and human-in-the-loop takeover without changing the role of the open-source package.

Use cases

• Complete repetitive forms and data-entry workflows, with a review step before consequential submissions.
• Extract structured records from sites that require navigation, filtering or interaction rather than a single HTTP request.
• Run browser-based QA against web applications, including navigation checks and issue collection.
• Automate scheduled or authenticated browser tasks while reusing approved session state.
• Add web navigation and interaction to a larger agent, internal tool or research pipeline.