Browser Use

What Is Browser Use?

Browser Use is a well-regarded open-source framework that makes it straightforward to build AI agents with full browser control. You give it a task in plain language and connect it to an AI model, and it drives a real browser to get the job done: opening pages, reading them, clicking, typing, and extracting what you asked for.

It has become a popular foundation for developers building browser automation that needs to handle dynamic, JavaScript-rendered pages, the kind traditional scrapers struggle with. If you have searched for “browser use ai agent” or “browser-use documentation,” this is the project you are looking for.

What Browser Use Is Best For

• Building browser agents: create AI agents that navigate and interact with real websites
• Web automation: automate repetitive web tasks that require a real, JavaScript-rendered page
• Data extraction: pull data from sites that simple scrapers cannot reach
• Form submission and login flows: automate multi-step forms and authenticated journeys
• Testing web apps: drive AI-based tests that interact with your application the way a person would

Browser Use Quick Start

The core library is a Python package. The typical shape of a project is: install it, set your model API key, define a task, and run an agent.

from browser_use import Agent, ChatAnthropic

agent = Agent(
    task="Find the current top post on Hacker News and return its title and URL",
    llm=ChatAnthropic(model="claude-opus-4-8"),
)
result = await agent.run()
print(result)

Browser Use handles the browser interaction, launching the browser, reading the page, and deciding what to click or type, while you define the task and the model. Package names and setup steps change between versions, so follow the official quick start in the documentation for the current commands.

Browser Use API and API Key

There are two things people mean by “Browser Use API”:

• Your model API key. Browser Use itself does not think; the AI model does. You supply a key for the model you want to use (for example an Anthropic, OpenAI, or Google key), usually as an environment variable, and Browser Use calls that model to decide each step.
• The hosted cloud API. Alongside the open-source library, there is a hosted cloud offering that runs browser agents for you and exposes its own API, so you do not have to manage browsers and infrastructure yourself. Features and pricing for the cloud change, so check the official site.

Keep your keys out of your code and out of task text. Store them in environment variables or a secrets manager, never paste them into a prompt.

Browser Use Models

Browser Use is model-agnostic, which is one of its strengths. You choose the model when you create the agent:

• Claude (Anthropic), including the current flagship models, for strong reasoning on multi-step tasks
• GPT-5 family (OpenAI)
• Gemini (Google)

More capable models tend to plan browser steps more reliably; cheaper models cost less per run but may need simpler tasks. Because browser tasks make many model calls, model choice is also the main lever on cost.

Browser Use vs a Ready-Made Browser Agent

Browser Use is a developer framework: you write code and bring a model. That is the right choice when you want control and are building something custom. If you want a browser agent you can just use, without writing code, a packaged product fits better.

• Choose Browser Use when you are a developer building custom automation and want full control.
• Choose a ready-made agent like Fellou or BrowserAct when you want browser automation without building it yourself.

For the concept behind all of these, see the browser agent glossary entry.

When To Use Browser Use

• You need to automate a website that has no usable API
• The pages are dynamic and JavaScript-heavy, so plain scraping fails
• You want an agent that can adapt to layout changes rather than a brittle script
• You are a developer comfortable in Python and managing a model API key

When Not To Use Browser Use

• A stable public API exists for the data or action you need; call that instead, it is faster and more reliable
• The task touches sensitive accounts, payments, or private data you should not hand to an automated agent
• You are not a developer and want a no-code tool; use a packaged browser agent
• You need guaranteed, exact behavior every time; AI-driven browsing is flexible but not perfectly deterministic

Safety and Permission Notes

Browser Use hands an AI model real control of a browser, so treat it like any automation that can act on your behalf:

• Least privilege. Do not log it into accounts it does not need. Prefer test accounts and sandboxes.
• No secrets in tasks. Never paste passwords, tokens, customer data, or other sensitive information into a task or prompt.
• Review actions. Watch what the agent does before you trust it with anything important, and keep a human in the loop for consequential steps.
• Respect sites’ terms. Automating a website may be against its terms of service. Check before you run at scale.

The computer-use agents guide and the AI agent permissions guide go deeper on running agents like this safely.

Honest Limitations

• Developer tool. It requires Python and programming knowledge.
• Model API costs. Browser tasks use many model calls; costs add up for complex workflows.
• Reliability on tricky sites. CAPTCHAs, anti-bot measures, and unusual layouts may need extra handling.
• Bring your own key. You need a model API key (Claude, OpenAI, Gemini, and others).

Alternatives Worth Knowing

• Fellou: a ready-to-use agentic browser for non-developers
• BrowserAct: browser automation aimed at less technical users
• Cua: computer-use agent infrastructure for developers
• LocalClicky: a local option for clicking through web tasks