# Browser Use > Source: https://aiwiki.ai/wiki/browser_use > Updated: 2026-10-28 > Fact-checked: 2026-11-05 > Categories: AI Agents, Developer Tools, Open Source AI > License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)" > Cite as: AI Wiki. "Browser Use." aiwiki.ai, 28 Oct 2026. https://aiwiki.ai/wiki/browser_use > From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution. Browser Use is an open-source Python library that lets [large language model](https://aiwiki.ai/wiki/large_language_model)s control a web browser, turning any LLM into an agent that can browse the web, click buttons, fill forms, and complete multi-step tasks online. It was released in late 2024 by Magnus Müller and Gregor Žunič, and the company behind it, Browser Use Inc., joined Y Combinator's Winter 2025 batch and raised a $17 million seed round led by Felicis Ventures in March 2025.[^3] The [GitHub](https://aiwiki.ai/wiki/github) repository grew from zero to over 50,000 stars within three months of launch, surpassed 100,000 stars by mid-2026, and had about 120,000 stars as of November 2026, making it one of the fastest-growing AI agent infrastructure projects on GitHub.[^1][^2][^22]
Browser Use logo
Browser Use logo, from the official website browser-use.com. Trademark of its owner.
Where proprietary alternatives like [OpenAI Operator](https://aiwiki.ai/wiki/openai_operator) and [Computer use](https://aiwiki.ai/wiki/computer_use) from [Anthropic](https://aiwiki.ai/wiki/anthropic) rely on closed APIs, Browser Use provides an open-source foundation that developers can inspect, modify, and self-host. It originally wrapped the [Playwright](https://aiwiki.ai/wiki/playwright) browser automation library in an LLM-driven control loop, exposing a simple Python interface for building web agents; in August 2025 (version 0.6.0) the project dropped Playwright entirely and switched to direct Chrome DevTools Protocol (CDP) control for faster execution and lower overhead.[^23][^24] Co-founder Magnus Müller summarized the core idea this way: "A lot of agents rely on vision-based systems and try and navigate websites through screenshots, and in [the] process, things break. We convert [websites] into something agents can understand."[^3] ## What is Browser Use used for? Browser Use is used to give AI agents reliable control of a real web browser so they can automate tasks that previously required a human at a keyboard: web research and price comparison, form filling and checkout flows, QA testing against staging environments, and data entry between SaaS tools that lack APIs. Because it represents pages as a structured element tree rather than as screenshots, it is cheaper and more robust than vision-first approaches on standard web pages, and it is distributed under the permissive [MIT license](https://aiwiki.ai/wiki/mit_license) so any developer can self-host it.[^1] Specific scenarios are detailed in the use cases section below.
"Browser Use Agent Tutorial: Automate Web Tasks with AI", official video from Browser Use, published January 7, 2026. Watch on YouTube.
## Who created Browser Use, and when was it released? Magnus Müller had been working on [web scraping](https://aiwiki.ai/wiki/web_scraping) tools for several years before meeting Gregor Žunič in 2024 while both were completing master's degrees in [data science](https://aiwiki.ai/wiki/data_science) at [ETH Zurich](https://aiwiki.ai/wiki/eth_zurich). The two began collaborating at ETH Zurich's Student Project House accelerator, a workspace designed for side projects. They built the first version of Browser Use in approximately five weeks and released it publicly in late 2024.[^7] The initial demo resonated with developers building AI agents who had struggled with brittle web automation. Müller and Žunič open sourced the project under the MIT license and the GitHub repository grew rapidly, attracting contributors and users building everything from personal automation scripts to commercial products.[^1] The speed of adoption caught the attention of [Y Combinator](https://aiwiki.ai/wiki/y_combinator), and the two founders joined the Winter 2025 batch in San Francisco. By early 2025, Browser Use had become a common dependency in [AI agent](https://aiwiki.ai/wiki/ai_agents) projects. The company announced its $17 million seed round in March 2025 and around the same time began rolling out Browser Use Cloud, a managed hosting product that lets teams use Browser Use without running their own browser infrastructure.[^3][^4] In March 2025, Chinese AI startup Butterfly Effect launched [Manus AI](https://aiwiki.ai/wiki/manus_ai), an autonomous agent that could complete complex multi-step tasks such as resume screening and stock analysis. [Manus](https://aiwiki.ai/wiki/manus) became a viral phenomenon, and Browser Use was one of the tools powering its web interaction capabilities. A post explaining this connection garnered over 2.4 million views on X, introducing Browser Use to a large audience outside the developer community.[^25] Through the second half of 2025 the project shifted from a single-purpose library into a layered product suite. The team shipped Profile Use for syncing real Chrome profiles in October 2025, achieved SOC 2 Type II compliance on November 4, 2025 after an audit by Accorp Partners covering July 17 to October 17, 2025, released a community Templates Library on November 13, 2025, added a hosted cloud [MCP server](https://aiwiki.ai/wiki/mcp_server) (the open-source library had shipped local MCP support in July 2025) and [Gemini 3](https://aiwiki.ai/wiki/gemini_3) support on November 21, 2025, launched the Skills API for plain-text automation workflows on December 4, 2025, and released open weights for BU-30B-A3B-Preview, its first open-source browser model, on December 16, 2025.[^5][^6][^17][^30] The pace accelerated in 2026. By February the engineering team had grown to roughly eight people, and Müller publicly noted he was spending less time writing code and more time in a CEO role.[^7] Browser Use Cloud restructured pricing on April 3, 2026 to add a permanent free tier, and on April 25, 2026 the company launched Browser Use Box (BUX), a 24/7 remote virtual machine bundling Claude Code with the Browser [Harness](https://aiwiki.ai/wiki/harness) and accessible through a web terminal, SSH, or a user-supplied Telegram bot.[^8][^9] ## How much funding has Browser Use raised? Browser Use announced a $17 million seed round on March 23, 2025. The round was led by [Felicis Ventures](https://aiwiki.ai/wiki/felicis_ventures) (partner Astasia Myers), with participation from A Capital, Nexus Venture Partners, Y Combinator, SV Angel, Pioneer Fund, Liquid2, and Paul Graham as an individual investor.[^3][^4][^18] At the time of the funding announcement, SiliconANGLE described Browser Use as a five-person team.[^18] More than 20 companies in the YC W25 batch were using Browser Use in their core workflows.[^3] The seed funding was directed toward building out Browser Use Cloud, expanding the team, and improving the reliability of the web agent framework. As of November 2026 no Series A round had been publicly reported. ## How does Browser Use work? Browser Use was originally built on top of [Playwright](https://aiwiki.ai/wiki/playwright), [Microsoft](https://aiwiki.ai/wiki/microsoft)'s open-source browser automation library, with Playwright handling low-level browser control and Browser Use adding an LLM reasoning layer using the Chrome DevTools Protocol (CDP). In August 2025 (version 0.6.0) the team removed Playwright entirely and rebuilt browser control on raw CDP, citing much faster element extraction, screenshots and default actions and more direct control over browser behavior.[^23][^24] The central architectural insight is how Browser Use represents web pages to the [language model](https://aiwiki.ai/wiki/language_model). Rather than sending raw HTML or screenshots to the LLM, it extracts and processes the page's [Document Object Model](https://aiwiki.ai/wiki/dom) (DOM) tree into a cleaned, structured representation that strips out irrelevant styling and metadata. This text-based format is considerably cheaper to process with a language model than pixel-based approaches, which must encode and decode full-resolution screenshots on every step. ### DOM extraction and the element tree The core component responsible for making web pages readable to LLMs is the `DomService`. When an agent loads a page, the process works as follows: 1. Multiple CDP calls gather the full browser state, including accessibility data, layout metrics, and interactive element locations. 2. A semantic tree is constructed from the page DOM, filtering out non-interactive and invisible elements to reduce noise. 3. Each interactive element receives a numeric index (buttons, input fields, links, dropdowns, checkboxes, and similar components). 4. This tree is serialized as a compact XML-style string included in the LLM's [context window](https://aiwiki.ai/wiki/context_window). The result looks something like: ``` [1] [2] [3] Forgot password? ``` The LLM reads this representation, decides which element to interact with, and returns a structured action command referencing elements by their index. This avoids the need to parse raw HTML or interpret screenshots for precise click targeting. ### Vision-based fallback For pages where DOM extraction is insufficient, Browser Use includes a screenshot-based fallback. When [Claude](https://aiwiki.ai/wiki/claude) Sonnet 4 or [Claude Opus 4](https://aiwiki.ai/wiki/claude_opus_4) models are used, coordinate clicking is automatically enabled, letting the agent click on arbitrary coordinates in a screenshot if the element tree does not contain a suitable target. This hybrid approach lets the library handle dynamic content, canvas elements, and pages that use non-standard rendering pipelines. ### Agent loop The `Agent` class orchestrates execution through an iterative loop: 1. The `MessageManager` constructs a prompt that includes the current browser state (element tree or screenshot), the task description, and the conversation history. 2. The prompt is sent to the configured LLM, which returns a structured action. 3. The `Tools` registry dispatches the action to the appropriate browser operation via CDP. 4. The browser executes the operation, and the new page state is observed. 5. The loop repeats until the agent calls the `done` action or a maximum step count is reached. An `ActionLoopDetector` monitors for repetitive patterns in the action history and can interrupt the loop to prevent the agent from cycling in place. An `AgentHistoryList` maintains a record of all previous steps, giving the LLM context about what has already been attempted. ### Action space Browser Use exposes a registry of high-level actions that the LLM can invoke. The built-in action set includes: | Action | Description | |--------|-------------| | `navigate` | Load a URL in the browser | | `click` | Click an element by its index | | `input` | Type text into an input field | | `scroll` | Scroll the page in a given direction | | `extract` | Run a separate LLM call against the page content to retrieve specific information | | `go_back` | Navigate to the previous page | | `wait` | Pause execution for a specified duration | | `screenshot` | Capture the current page state as an image | | `select_dropdown` | Select an option from a dropdown menu | | `upload_file` | Upload a file to a file input | | `evaluate` | Execute arbitrary [JavaScript](https://aiwiki.ai/wiki/javascript) on the page | | `done` | Signal task completion and return a result | Developers can register custom actions by decorating Python functions with the `@action` decorator and passing them to the agent at initialization. Custom actions give agents access to domain-specific operations such as writing to a database, calling an external API, or reading from the local file system. The `extract` action is a notable design choice. Rather than asking the main LLM to reason about large amounts of page text, it spawns a separate, lightweight LLM call against the page's markdown content to pull out only the relevant data, keeping the main context window focused on decision-making. ### Skills Released on December 4, 2025, the Skills API turns a plain-text description of an automation goal into a reusable API endpoint: Browser Use generates parameter and output schemas for the skill, which can then be run through the API with typed inputs.[^6] ## How do you install and use Browser Use? Installing Browser Use requires Python 3.11 or higher. The package is available on PyPI: ``` uv add browser-use ``` A minimal agent looks like this: ```python from browser_use import Agent, ChatOpenAI import asyncio agent = Agent( task="Find the price of a MacBook Pro on apple.com", llm=ChatOpenAI(model="gpt-4o"), ) asyncio.run(agent.run()) ``` The `Agent` class accepts one of Browser Use's native chat model wrappers (the library replaced its [LangChain](https://aiwiki.ai/wiki/langchain) integration with its own LLM classes in June 2025), a task description in plain English, and optional configuration for browser settings, maximum number of steps, and callbacks. The agent runs asynchronously and returns a structured result when it completes or fails. More complex workflows can chain multiple agents, define custom actions that extend the agent's capabilities beyond the built-in browser actions, or integrate Browser Use into larger agent systems. The library exposes hooks for observing agent state at each step, which makes it possible to build monitoring dashboards, inject human-in-the-loop approvals, or log detailed traces for debugging. ## Which LLMs does Browser Use support? Browser Use is model-agnostic. Since version 0.3.3 (June 2025) it uses its own native model wrappers rather than LangChain's abstraction layer.[^17] The library maintains wrapper classes for a wide range of providers: | Provider | Notes | |----------|-------| | Browser Use Cloud (`ChatBrowserUse`) | Proprietary model optimized for browser tasks; variants include `bu-1-0` (alias `bu-latest`) and `bu-2-0` (released January 27, 2026) | | [OpenAI](https://aiwiki.ai/wiki/openai) (`ChatOpenAI`) | [GPT-4o](https://aiwiki.ai/wiki/gpt_4o), [GPT-4.1](https://aiwiki.ai/wiki/gpt-4.1), and compatible models | | [Anthropic](https://aiwiki.ai/wiki/anthropic) (`ChatAnthropic`) | Claude Sonnet and Opus; coordinate clicking auto-enabled for Sonnet 4 and Opus 4 | | [Google](https://aiwiki.ai/wiki/google) (`ChatGoogle`) | [Gemini](https://aiwiki.ai/wiki/gemini) 2.5 Flash and Pro; Vertex AI and Gemma supported; [Gemini 3](https://aiwiki.ai/wiki/gemini_3) added November 21, 2025 | | Azure [OpenAI](https://aiwiki.ai/wiki/openai) | Azure-hosted GPT models | | AWS Bedrock | Claude and other Bedrock-hosted models | | [DeepSeek](https://aiwiki.ai/wiki/deepseek) | [DeepSeek-V3](https://aiwiki.ai/wiki/deepseek_v3), [DeepSeek-R1](https://aiwiki.ai/wiki/deepseek_r1) | | Mistral | Mistral and Mixtral models | | Groq | Llama models served via Groq's inference API | | [Ollama](https://aiwiki.ai/wiki/ollama) | Local models | | [OpenRouter](https://aiwiki.ai/wiki/openrouter) | Access to 300+ models through a single endpoint | | [LiteLLM](https://aiwiki.ai/wiki/litellm) | Any LiteLLM-compatible provider string | The `ChatBrowserUse` model is a proprietary model built by Browser Use for browser automation tasks. On December 16, 2025 Browser Use released open weights for BU-30B-A3B-Preview, a model with 30 billion total and 3 billion active parameters that it said completes about 200 tasks per dollar, four times the cost efficiency of BU 1.0; Browser Use Cloud does not serve that open model.[^6] BU 2.0 followed on January 27, 2026, reporting a 12% relative accuracy gain over BU 1.0 (74.7% to 83.3%, about 8.6 points, on Browser Use's benchmark) at similar latency (about 62 seconds per task versus 58 seconds for BU 1.0). The team described BU 2.0 as matching [Claude Opus 4.5](https://aiwiki.ai/wiki/claude_opus_4_5) accuracy while running about 40% faster.[^11] For cost-sensitive applications, smaller models like [GPT-4o mini](https://aiwiki.ai/wiki/gpt_4o_mini) and [Gemini 2.0 Flash](https://aiwiki.ai/wiki/gemini_2_0_flash) work reasonably well on straightforward tasks, while complex multi-step workflows typically benefit from more capable models. Vision-language models are preferable for pages with complex visual layouts that the DOM representation alone does not capture accurately. ## What is Browser Use Cloud? In addition to the open-source library, Browser Use operates a managed cloud platform at browser-use.com. The cloud platform provides managed browser infrastructure so developers do not need to run Chrome locally. Browser Use Cloud offers several deployment modes. In Browser Harness mode, teams use their own agent code (the open-source library) but connect to cloud-hosted browsers rather than spinning up local Chromium instances. In Hosted V3 Agent mode (introduced with the experimental Agent API and SDK 3.0 on February 25, 2026), a fully managed agent accepts a task description and handles everything from browser control to LLM reasoning without requiring any agent code from the user.[^6] Later in 2026 the company made its Browser Harness (v4) agent API public (July 17, 2026), added scheduled V4 agents (August 2026), and on October 30, 2026 opened its BU Ultrafast model to every Browser Use Cloud plan in the US.[^6] Key features of Browser Use Cloud include stealth browsers with fingerprinting bypass across 195+ countries, built-in CAPTCHA solving (including an hCaptcha solver added with CLI 2.0 and improved reCAPTCHA accuracy), managed proxy pools, parallel session execution, and persistent browser profiles for sites requiring authentication. November 4, 2025 saw the company complete SOC 2 Type II certification, audited by Accorp Partners for July 17 to October 17, 2025.[^5][^30] ### Browser Use Box (BUX) Launched on April 25, 2026, Browser Use Box is a remote virtual machine that comes pre-loaded with [Claude Code](https://aiwiki.ai/wiki/claude_code) and the Browser Harness. Spin-up takes 30 to 60 seconds from a baked AMI, browser sessions automatically rotate every 240 minutes, and the VM runs under a locked-down IAM role with zero AWS permissions. Users can attach a Telegram bot token created via @BotFather to control the agent from a phone, SSH directly into the box, or use the cloud.browser-use.com/bux web terminal. The launch announcement positioned BUX as a workaround for complex dashboards such as [Microsoft Azure](https://aiwiki.ai/wiki/azure) and Google Workspace that pure browser agents had historically struggled to operate reliably.[^9] ### How much does Browser Use Cloud cost? Browser Use Cloud reorganized its pricing on April 3, 2026 around a free starter tier and per-credit usage on top of subscriptions, and cut proxy data from $10 to $5 per GB.[^8] A June 2026 snapshot of the pricing page listed browser sessions at $0.02 per hour, proxy bandwidth at $5 per GB, the V3 Agent at 1.2 times provider token rates (for example, [Claude Sonnet 4.6](https://aiwiki.ai/wiki/claude_sonnet_4_6) at $3.60 input / $18.00 output per million tokens), and Browser Use Box at $1, $2 or $4 per day depending on the compute tier.[^31] As of November 2026 the company had moved to pure pay-as-you-go billing with no subscription plans offered: cloud browsers cost $0.02 per hour plus $5 per GB of residential proxy traffic, web agents are billed at model token rates plus a 20% fee, concurrency rises with total payments (10 sessions at $0, up to 1,000 at $25,000), and eligible new signups get $1 in free credit.[^12] The table below shows the subscription plans in the June 2026 snapshot, now kept only as legacy plans for existing subscribers.[^31] | Plan | Monthly price | Credits included | Concurrent sessions | |------|---------------|-----------------|---------------------| | Free | $0 | Free browsers (10 agent tasks per month) | 3 | | Dev | $29 | $29 | 25 | | Business | $299 | $299 | 200 | | Scaleup | $999 | $999 | 500 | | Enterprise | Custom | Negotiated | Negotiated | Bring-your-own-key is supported: customers using their own Anthropic, [OpenAI](https://aiwiki.ai/wiki/openai), or [Google](https://aiwiki.ai/wiki/google) keys pay their provider directly, and Browser Use charges 20% of published token rates plus browser time and traffic.[^12] ## How does Browser Use compare with OpenAI Operator and Anthropic Computer Use? Browser Use occupies a different position in the market from the two most prominent proprietary alternatives, OpenAI Operator and Anthropic's Computer use. The three approaches differ significantly in architecture, cost structure, and intended audience. | Feature | Browser Use | OpenAI Operator | Anthropic Computer Use | |---------|------------|-----------------|------------------------| | Open source | Yes (MIT) | No | No | | Self-hostable | Yes | No | No | | Interaction approach | DOM element tree (+ vision fallback) | Screenshots (vision-first) | Screenshots (vision-first) | | Model support | Any LLM (15+ providers) | Proprietary CUA model | Claude only | | Desktop app control | No | No | Yes | | Developer customization | High (custom actions, model choice) | Low in Operator itself; the CUA model was offered as a Responses API tool from March 2025 | Limited | | [WebVoyager](https://aiwiki.ai/wiki/webvoyager) accuracy | 89.1% (December 2024 figure) | 87% (CUA, January 2025) | ~56% (prior computer-use figure cited by OpenAI, January 2025) | | Pricing | Free self-hosted; pay-as-you-go cloud | Launched for ChatGPT Pro ($200/mo); folded into ChatGPT agent in July 2025 | Token-based via Claude API | OpenAI Operator was a standalone consumer-facing agent product that ran in its own sandboxed browser; in July 2025 OpenAI folded its capabilities into ChatGPT agent and said the Operator site would be sunset a few weeks later.[^28] It is designed for non-technical users and offers a polished interface with a takeover mode for entering sensitive credentials. Operator itself had no developer API, although OpenAI offered the same Computer-Using Agent model to developers through the Responses API's computer use tool from March 2025.[^29] [Anthropic Computer Use](https://aiwiki.ai/wiki/anthropic_computer_use) works by giving Claude the ability to view screenshots of a desktop environment and issue mouse and keyboard commands. Because it operates at the pixel level, it can control any desktop application, not just web pages. However, this pixel-level approach is slower, more expensive in tokens, and less reliable on web-specific tasks than DOM-based methods. Browser Use targets developers. The open-source library is free, works with any LLM, and can be customized extensively through the action registry. The trade-off is that it requires Python experience and cannot control non-browser desktop applications. Manus AI used Browser Use as its web interaction layer, demonstrating that the library can serve as infrastructure inside a larger agentic system. ## How accurate is Browser Use on the WebVoyager benchmark? Browser Use published a technical report presenting results on the [WebVoyager](https://aiwiki.ai/wiki/webvoyager) benchmark, a widely used evaluation suite for web agents covering tasks across diverse websites including Amazon, Google Flights, GitHub, [Hugging Face](https://aiwiki.ai/wiki/hugging_face), Apple, and Booking.com.[^13] On a modified WebVoyager set of 586 web tasks (the report says 55 tasks were removed, chiefly ones made impossible by outdated data, and lists [Cloudflare](https://aiwiki.ai/wiki/cloudflare) blocks and ambiguous prompts as further caveats), Browser Use achieved an 89.1% success rate using [GPT-4o](https://aiwiki.ai/wiki/gpt_4o) as the primary model. OpenAI later reported 87% on WebVoyager for its Computer-Using Agent, the model behind Operator, in January 2025, though not on Browser Use's modified task set.[^27] Results by domain: | Domain | Success rate | |--------|-------------| | [Hugging Face](https://aiwiki.ai/wiki/hugging_face) | 100% | | Google Flights | 95% | | Amazon | 92% | | [GitHub](https://aiwiki.ai/wiki/github) | 92% | | Apple | 91% | | Booking.com | 80% | The Magnitude web agent subsequently reported a 93.9% success rate on WebVoyager using a different technical approach, though benchmark conditions, model versions, and evaluation methodologies vary substantially across implementations.[^14] As of November 2026, BenchLM's WebVoyager table, built mostly from provider self-reports, was led by Microsoft's Fara1.5-27B at 89.3%.[^15] The WebVoyager benchmark has known limitations and is best treated as a directional indicator rather than a definitive ranking; the report itself notes that the team manually reviewed and corrected evaluations the default WebVoyager judge had marked unknown or failed, and that it removed 55 tasks from the standard set.[^13] ## What can you build with Browser Use? (use cases) Developers use Browser Use across a wide range of automation scenarios. Web research and data collection agents visit multiple websites to compare prices, gather product specifications, collect publicly available data, or aggregate information across sources. Because the library works with authenticated sessions, it can access data on sites that would block a traditional scraper. Form automation covers filling out and submitting web forms, including multi-step workflows like job applications, registration flows, and checkout processes. The agent reads the form structure dynamically, reducing maintenance overhead compared to brittle CSS-selector-based scripts. CRM and SaaS data entry is another common use case. Companies transfer data between SaaS platforms that lack official API integrations; the agent logs into each platform using a stored browser profile and performs the data transfer through the user interface. The Browser Use Box launch specifically highlighted complex enterprise dashboards such as Microsoft Azure and Google Workspace as a category that benefits from a long-running remote VM rather than a stateless cloud session.[^9] QA teams write natural-language test scripts that the agent executes against staging environments. Unlike Selenium or Playwright scripts, the agent can handle dynamic layouts that change between releases without requiring selector updates. Shopping agents compare product availability, track price changes, add items to carts, and complete purchases when prices hit a target threshold. One widely shared demo showed an agent filling out job applications by reading job requirements, matching them to a resume, and completing application forms across dozens of employer sites. ## What are the limitations of Browser Use? Browser Use has several constraints worth considering. The library controls a web browser only. It cannot interact with native desktop applications, system dialogs, or operating system interfaces. For those use cases, Computer Use or screen-recording-based agents are more appropriate. Agent performance depends on the quality of the underlying LLM. On complex tasks with many steps, smaller or less capable models may make incorrect decisions that cause cascading failures. The 89.1% WebVoyager result was achieved with GPT-4o; smaller models score lower. Some sites aggressively detect automated browsers through fingerprinting, CAPTCHA challenges, or behavior analysis. The open-source library does not include stealth features by default; the cloud platform's stealth browsers are required for consistently accessing these sites. Sites that load content asynchronously may present an empty or partial element tree if the agent acts before [JavaScript](https://aiwiki.ai/wiki/javascript) finishes rendering. The library includes a `wait` action and some automatic stabilization, but highly dynamic pages can still cause premature action on incomplete states. On pages with very large element trees, the serialized representation can consume a significant portion of the LLM's context window, leaving less room for task history and reasoning. The `DomService` applies filtering heuristics to reduce tree size, but complex pages with many interactive elements can still produce large inputs. The benchmark methodology also raises comparability questions. The team removed 55 tasks from the WebVoyager evaluation set, which means the reported 89.1% figure is not directly comparable to implementations that run the original unmodified benchmark. ### LiteLLM supply chain incident (March 2026) On March 24, 2026, the upstream litellm Python package was hit by a supply chain attack in which versions 1.82.7 and 1.82.8 were published with malicious code that exfiltrated cloud and CI/CD credentials. The compromised packages were live on PyPI for approximately 40 minutes before being quarantined. Because litellm had been added as a core dependency of Browser Use in version 0.12.3 the day before, the project shipped emergency releases on March 24 (`0.12.4`, pinning a known-good litellm version) and March 25 (`0.12.5`, removing litellm from core dependencies entirely and exposing `ChatLiteLLM` only as an optional install).[^16][^17] Users who installed Browser Use through the broader litellm dependency chain during the exposure window were advised to rotate API keys and audit their environments. ## How popular is Browser Use? Browser Use drew immediate attention from the developer community when it launched in late 2024. The GitHub repository grew to 50,000 stars in roughly three months, with 15,000 or more developers actively building on the library. By mid-2026 the count had passed 100,000, and it stood at about 120,000 in November 2026, ranking it among the most-starred AI agent repositories on GitHub.[^1][^2][^22] TechCrunch covered the $17 million seed round in March 2025, describing Browser Use as "the tool making it easier for AI agents to navigate websites."[^3] The Runa Capital ROSS Index for Q1 2025 ranked Browser Use second among the fastest-growing open-source startups.[^26] SiliconANGLE also covered the funding, and InfoWorld had profiled the library in January 2025.[^18][^19] Within the YC W25 batch, more than 20 companies adopted Browser Use for core product functionality, an unusually rapid rate of upstream integration for a library that was only a few months old at Demo Day.[^3] Some developers raised security concerns about running Chrome in remote debugging mode without sandboxing, which opens a CDP port that could be exploited by malicious pages via XSS attacks. The Browser Use team addressed this in documentation by recommending that production deployments use cloud-hosted browsers or configure proper network isolation. The litellm incident in March 2026 sharpened these concerns and prompted the team to publicly tighten the project's dependency posture by removing litellm from the core install.[^17] ## See also - [OpenAI Operator](https://aiwiki.ai/wiki/openai_operator) - [Computer use](https://aiwiki.ai/wiki/computer_use) - [Manus AI](https://aiwiki.ai/wiki/manus_ai) - [AI agent](https://aiwiki.ai/wiki/ai_agent) - [LangChain](https://aiwiki.ai/wiki/langchain) - [Playwright](https://aiwiki.ai/wiki/playwright) - [Claude Code](https://aiwiki.ai/wiki/claude_code) - [Web scraping](https://aiwiki.ai/wiki/web_scraping) - [Robotic process automation](https://aiwiki.ai/wiki/robotic_process_automation) - [Y Combinator](https://aiwiki.ai/wiki/y_combinator) ## References [^1]: Browser Use GitHub Repository. https://github.com/browser-use/browser-use [^2]: "Browser Use: Leading open-source web agent project with 50k stars in 3 months." Y Combinator. https://www.ycombinator.com/companies/browser-use [^3]: "Browser Use, the tool making it easier for AI 'agents' to navigate websites, raises $17M." TechCrunch, March 23, 2025. https://techcrunch.com/2025/03/23/browser-use-the-tool-making-it-easier-for-ai-agents-to-navigate-websites-raises-17m/ [^4]: "We Raised $17M to Build the Future of Web for Agents." Browser Use Blog, March 23, 2025. https://browser-use.com/posts/seed-round [^5]: "SOC 2 Type II Compliance." Browser Use Changelog, November 4, 2025. https://browser-use.com/changelog/4-11-2025 [^6]: Browser Use Changelog index. https://browser-use.com/changelog [^7]: "Founder Story: Magnus Müller of Browser Use." Frederick.ai. https://www.frederick.ai/blog/magnus-muller-browser-use [^8]: "Free Tier & Pricing Update." Browser Use Changelog, April 3, 2026. https://browser-use.com/changelog [^9]: "BUX: Your 24/7 Remote Agent with Browser Harness." Browser Use Blog, April 25, 2026. https://browser-use.com/posts/bux-launch-blog [^10]: "CLI 2.0 + Weekly Update." Browser Use Changelog, March 22, 2026. https://browser-use.com/changelog/22-3-2026 [^11]: "Browser Use Model - BU 2.0." Browser Use Changelog, January 27, 2026. https://browser-use.com/changelog/27-1-2026 [^12]: Browser Use Cloud Pricing. https://browser-use.com/pricing [^13]: "Browser Use = state of the art Web Agent." Browser Use Technical Report. https://browser-use.com/posts/sota-technical-report [^14]: Magnitude WebVoyager submission. https://github.com/magnitudedev/webvoyager [^15]: "WebVoyager Benchmark 2026: 3 model averages." BenchLM.ai. https://benchlm.ai/benchmarks/webVoyager [^16]: "Security Update: Suspected Supply Chain Incident." LiteLLM, March 2026. https://docs.litellm.ai/blog/security-update-march-2026 [^17]: Browser Use Releases. https://github.com/browser-use/browser-use/releases [^18]: "Browser Use raises $17M to help steer AI agents through the internet." SiliconANGLE, March 23, 2025. https://siliconangle.com/2025/03/23/browser-use-raises-17m-help-steer-ai-agents-internet/ [^19]: "Browser Use: An open-source AI agent to automate web-based tasks." InfoWorld. https://www.infoworld.com/article/3812644/browser-use-an-open-source-ai-agent-to-automate-web-based-tasks.html [^20]: browser-use package on PyPI. https://pypi.org/project/browser-use/ [^21]: "Building Browser Use, going through YC & raising a $17M Seed. Meet Magnus." Ambitious x Driven. https://www.ambitiousxdriven.com/p/building-browser-use-going-through [^22]: Browser Use GitHub Repository (star count, mid-2026: over 100,000 stars). https://github.com/browser-use/browser-use [^23]: "We switched from Playwright to CDP." Browser Use Changelog, August 19, 2025. https://browser-use.com/changelog/19-8-2025 [^24]: "Closer to the Metal: Leaving Playwright for CDP." Browser Use Blog, August 20, 2025. https://browser-use.com/posts/playwright-to-cdp [^25]: "Browser Use, one of the tools powering Manus, is also going viral." TechCrunch, March 12, 2025. https://techcrunch.com/2025/03/12/browser-use-one-of-the-tools-powering-manus-is-also-going-viral/ [^26]: Runa Capital. "ROSS Index 2025 Q1." https://github.com/RunaCapital/ROSS-Index/blob/main/2025/Q1/ross-index-2025-q1.md [^27]: "Computer-Using Agent." OpenAI, January 23, 2025. https://web.archive.org/web/2026/https://openai.com/index/computer-using-agent/ [^28]: "Introducing ChatGPT agent." OpenAI, July 17, 2025. http://web.archive.org/web/20250801053448/https://openai.com/index/introducing-chatgpt-agent/ [^29]: "New tools for building agents." OpenAI, March 11, 2025. http://web.archive.org/web/20250402104613/https://openai.com/index/new-tools-for-building-agents/ [^30]: "Security." Browser Use. https://browser-use.com/security [^31]: Browser Use Cloud Pricing (archived June 6, 2026). http://web.archive.org/web/20260606064244/https://browser-use.com/pricing