BrowserUse
title: browser-use type: tool created: 2026-05-26 last_updated: 2026-05-26 related: ["radar/tools/Firecrawl", "radar/tools/Fabric", "radar/tools/LlmCli"] sources: ["https://github.com/gregpr07/browser-use"] radar_quadrant: Tools radar_ring: Assess radar_position: inner
Open-source Python library connecting any LLM to a real browser via Playwright. The model observes the browser's current state and emits structured actions (click, type, scroll, navigate, extract) to complete natural-language tasks on any website, including JavaScript-rendered pages and sites requiring authentication.
browser-use exposes the browser state to the LLM as an observation: a simplified DOM representation, optional screenshot, and current URL. The LLM responds with a structured action. The library executes the action, captures the updated state, and loops until the task is complete or the model signals done.
The loop is managed by browser-use; the caller provides a task description and an LLM backend. Model selection is via LiteLLM, supporting Claude, GPT-4o, Gemini, and local models via Ollama.
- Automating multi-step workflows on sites with no API
- Extracting data from sites that block static scrapers or require login
- Building web-browsing agents as components in larger LLM pipelines
- Filling forms, navigating paginated results, downloading reports
browser-use occupies the same space as finic (seen in this radar session) but with different design priorities. browser-use is a Python library for embedding in agent workflows; finic is a service-oriented framework. browser-use's LiteLLM integration makes model swapping trivial, which is useful for cost/quality trade-offs across different task types. Both are more resilient to layout changes than traditional Playwright scripts because the LLM navigates by semantics rather than CSS selectors.
browser-use sits in the Assess ring of the Tools quadrant, at inner position. First studied via GitHub (2024-11-07). Web automation for LLM agents is a recurring need with no clean solution: static scrapers break on JS-rendered pages, traditional Playwright scripts break on layout changes, and hosted scraping APIs add cost and latency. browser-use addresses this with a Python-native, model-agnostic library that fits directly into existing LLM agent code. Complements Firecrawl (site-scale static ingestion) for cases requiring interaction or authentication. Remaining gate before Trial is a working agent completing a multi-step web task with measurable success rate over 20+ runs.
