{"id":3557,"date":"2025-02-05T14:56:27","date_gmt":"2025-02-05T13:56:27","guid":{"rendered":"https:\/\/blog.tomayac.com\/2025\/02\/05\/testing-browser-use-a-scriptable-ai-browser-agent\/"},"modified":"2025-02-05T14:56:27","modified_gmt":"2025-02-05T13:56:27","slug":"testing-browser-use-a-scriptable-ai-browser-agent","status":"publish","type":"post","link":"https:\/\/tomayac.com\/wordpress\/2025\/02\/05\/testing-browser-use-a-scriptable-ai-browser-agent\/","title":{"rendered":"Testing browser-use, a scriptable AI browser agent"},"content":{"rendered":"\n\t\t\t

I'm not a big LinkedIn user, but the other day, my Google colleague Franziska\nHinkelmann\nposted something<\/a>\nabout a project called\nbrowser-use<\/a> that caught my eye:<\/p>\n

\n

Got low stakes repetitive tasks in the browser? Playwright + LLMs (Gemini 2.0)\nto the rescue! Super easy to make somebody else cough<\/em> agents cough<\/em> do the\nwork for you, especially if you have to repeat a task for many rows in a\nGoogle Sheet.<\/p>\n<\/blockquote>\n

After seeing her demo, I went and tried it out myself. Here are the steps that\nworked for me on macOS:<\/p>\n

    \n
  1. \n

    Install uv<\/code> following their\ninstallation instructions<\/a>.\n(The usual caveat of first checking the source code before pasting anything\nin the Terminal applies.)<\/p>\n

    curl<\/span> -LsSf<\/span> https:\/\/astral.sh\/uv\/install.sh |<\/span> less<\/span><\/code><\/pre>\n<\/li>\n
  2. \n
    Create a new Python environment and activate it. This is from browser-use's\nquickstart<\/a> instructions.<\/p>\n
    uv venv --python<\/span> 3.11<\/span>\nsource<\/span> .venv\/bin\/activate<\/code><\/pre>\n<\/li>\n
  3. \n
    Install the dependencies and Playwright<\/a>.<\/p>\n
    uv pip install<\/span> browser-use\nplaywright install<\/span><\/code><\/pre>\n<\/li>\n
  4. \n
    Create a .env<\/code> file and add your OpenAI API key in the form\nOPENAI_API_KEY=abc123<\/code>.<\/p>\n<\/li>\n
  5. \n
    Create an agent.py<\/code> file with the source code of your agent. Here's the one\nI tried. As you can see, I'm tasking the agent with the following job: "Go\nto developer.chrome.com<\/a> and find out what built-in AI APIs Chrome supports".<\/em><\/p>\n
    from<\/span> langchain_openai import<\/span> ChatOpenAI\nfrom<\/span> browser_use import<\/span> Agent\nimport<\/span> asyncio\nfrom<\/span> dotenv import<\/span> load_dotenv\nload_dotenv(<\/span>)<\/span>\n\nasync<\/span> def<\/span> main<\/span>(<\/span>)<\/span>:<\/span>\n    agent =<\/span> Agent(<\/span>\n        task=<\/span>\"Go to developer.chrome.com and find out what built-in AI APIs Chrome supports.\"<\/span>,<\/span>\n        llm=<\/span>ChatOpenAI(<\/span>model=<\/span>\"gpt-4o\"<\/span>)<\/span>,<\/span>\n    )<\/span>\n    result =<\/span> await<\/span> agent.<\/span>run(<\/span>)<\/span>\n    print<\/span>(<\/span>result)<\/span>\n\nasyncio.<\/span>run(<\/span>main(<\/span>)<\/span>)<\/span><\/code><\/pre>\n<\/li>\n<\/ol>\n
    The coolest thing is that you can watch the agent do its job in real time and\nsee the reasoning steps in the command line.<\/p>\n
    \"Headless<\/p>\n
    Here's an excerpt of the logs.<\/p>\n
    INFO     [<\/span>agent]<\/span> \ud83d\ude80 Starting task: Go to developer.chrome.com and find<\/span> out what built-in AI APIs Chrome supports.\nINFO     [<\/span>agent]<\/span> \ud83d\udccd Step 1<\/span>\nINFO     [<\/span>agent]<\/span> \ud83e\udd37 Eval: Unknown - It's the starting state.\nINFO     [agent] \ud83e\udde0 Memory: Start searching for built-in AI APIs supported by Chrome on developer.chrome.com.\nINFO     [agent] \ud83c\udfaf Next goal: Navigate to developer.chrome.com\nINFO     [agent] \ud83d\udee0\ufe0f  Action 1\/1: {\"go_to_url\":{\"url\":\"https:\/\/developer.chrome.com\"}}\nINFO     [controller] \ud83d\udd17  Navigated to https:\/\/developer.chrome.com\nINFO     [agent] \ud83d\udccd Step 2\nINFO     [agent] \ud83d\udc4d Eval: Success - Navigated to the correct URL.\nINFO     [agent] \ud83e\udde0 Memory: Visited developer.chrome.com. Need to search for information on built-in AI APIs supported by Chrome.\nINFO     [agent] \ud83c\udfaf Next goal: Use search input to find information on AI APIs.\nINFO     [agent] \ud83d\udee0\ufe0f  Action 1\/1: {\"click_element\":{\"index\":9}}\nINFO     [controller] \ud83d\uddb1\ufe0f  Clicked button with index 9:\nINFO     [agent] \ud83d\udccd Step 3\nINFO     [agent] \ud83d\udc4d Eval: Success - The search bar was clicked but no input was entered yet.\nINFO     [agent] \ud83e\udde0 Memory: Need to input search query about AI APIs in Chrome using the search bar.\nINFO     [agent] \ud83c\udfaf Next goal: Enter '<\/span>AI APIs' in<\/span> the search input to find<\/span> relevant information.\nINFO     [<\/span>agent]<\/span> \ud83d\udee0\ufe0f  Action 1<\/span>\/1: {<\/span>\"input_text\"<\/span>:{<\/span>\"index\"<\/span>:4,\"text\"<\/span>:<\/span>\"AI APIs\"<\/span>}<\/span>}<\/span>\nINFO     [<\/span>controller]<\/span> \u2328\ufe0f  Input AI APIs into index 4<\/span>\n[<\/span>\u2026]<\/span><\/code><\/pre>\n
    At the end of the log messages is the agent's full response to your task,\nformatted in Markdown. FWIW, the answer is exactly correct:<\/p>\n
    \n
    Chrome supports several built-in AI APIs, including:<\/p>\n
      \n
    1. Translator API<\/strong> - Available in origin trial for early preview\nparticipants. Use cases include on-demand translation and supporting\nmulti-language communication.<\/li>\n
    2. Language Detector API<\/strong> - Available in Chrome origin trial. It helps\ndetect languages without user input and can label texts for better screen\nreader pronunciation.<\/li>\n
    3. Summarizer API<\/strong> - Allows for local experimentation to preview program\nparticipants. It can summarize meeting transcripts, articles, and forum\nquestions.<\/li>\n
    4. Writer and Rewriter APIs<\/strong> - Experimental status in early preview\nprogram, used for creating and refining text content.<\/li>\n
    5. Prompt API<\/strong> - Allows natural language requests to Gemini Nano in Chrome,\nin an experimental early stage.<\/li>\n<\/ol>\n
      Visit developer.chrome.com<\/a> for complete details and participation in early\ntrials.<\/p>\n<\/blockquote>\n
      It's pretty wild what this scriptable agent is capable of doing today. Be sure\nto check out some of the other\ndemos<\/a> and\nalso try the browser-use web-ui<\/a>, which\nadds a nice UI on top.<\/p>\n\n\t\t\t
      \n\t\t\t\t\"Thomas\n\t\t\t\tThis post appeared first on https:\/\/blog.tomayac.com\/2025\/02\/05\/testing-browser-use-a-scriptable-ai-browser-agent\/<\/a>.\n\t\t\t<\/p>\n\t\t","protected":false},"excerpt":{"rendered":"
      \t\t\tI’m not a big LinkedIn user, but the other day, my Google colleague Franziska
      \nHinkelmann
      \nposted something
      \nabout a project called
      \nbrowser-use that caught my eye:<\/p>\n
      Got low stakes repetitive tasks in the browser? Playwright + LLMs (Gemini 2.0)
      \nto the …<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"intlwemo_wallet_address":"","footnotes":""},"categories":[],"tags":[],"class_list":["post-3557","post","type-post","status-publish","format-standard","hentry"],"_links":{"self":[{"href":"https:\/\/tomayac.com\/wordpress\/wp-json\/wp\/v2\/posts\/3557","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/tomayac.com\/wordpress\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/tomayac.com\/wordpress\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/tomayac.com\/wordpress\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/tomayac.com\/wordpress\/wp-json\/wp\/v2\/comments?post=3557"}],"version-history":[{"count":1,"href":"https:\/\/tomayac.com\/wordpress\/wp-json\/wp\/v2\/posts\/3557\/revisions"}],"predecessor-version":[{"id":3558,"href":"https:\/\/tomayac.com\/wordpress\/wp-json\/wp\/v2\/posts\/3557\/revisions\/3558"}],"wp:attachment":[{"href":"https:\/\/tomayac.com\/wordpress\/wp-json\/wp\/v2\/media?parent=3557"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/tomayac.com\/wordpress\/wp-json\/wp\/v2\/categories?post=3557"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/tomayac.com\/wordpress\/wp-json\/wp\/v2\/tags?post=3557"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}