Browser Use

🌐 Make websites accessible for AI agents. Automate tasks online with ease.

Last verified:

Visit Browser Use

What is Browser Use?

Browser Use is an AI browser automation platform that enables AI agents to interact with and navigate websites more effectively than traditional vision-based approaches. Instead of trying to

Browser Use pricing

Pricing model: Freemium

Browser Use offers a Free plan at $0/month with 3 concurrent sessions and free browsers (no monthly credits). Dev plan is $29/month with $29 monthly credits, 25 concurrent sessions, 5 team members, 5 browser profiles, premium proxy pool, plus features like Bring Your Own Key, manual top-ups, auto recharge, scheduled jobs, priority support, advanced stealth, CAPTCHA solving, and webhook events. Business plan is $299/month with $400 monthly credits ($299 + $101 bonus), 200 concurrent sessions, unlimited team members, unlimited browser profiles, and all Dev features plus dedicated support. Scaleup plan is $999/month with $1,400 monthly credits ($999 + $401 bonus), 500 concurrent sessions, unlimited team members and profiles. Enterprise offers custom pricing with annual credit pool, negotiable credits, negotiable concurrent sessions, SLAs, data retention terms, and dedicated support. Browser sessions cost $0.06/hour and proxy bandwidth is $5/GB. V3 Agent uses token-based pricing at 1.2x provider rates (GPT-5.4 Mini: $0.90/$5.40 per 1M tokens, Claude Sonnet 4.6: $3.60/$18.00, Claude Opus 4.6/4.7: $6.00/$30.00). Browser Use Box is billed hourly: small at $1/day (2 vCPU, 4GB, 20GB), medium at $2/day (2 vCPU, 8GB, 50GB), large at $4/day (4 vCPU, 16GB, 100GB).

Browser Use pros

  • State-of-the-art browser agent with 97% on Online-Mind2Web benchmark (highest score ever)
  • 79k+ GitHub stars for open-source library proving community trust
  • Advanced stealth browsers that pass CreepJS, BrowserLeaks, and fingerprint detectors
  • CAPTCHA solving included by default with no configuration needed
  • Residential proxies across 195+ countries with rotating undetected IPs
  • Cloudflare and anti-bot bypass works out of the box
  • Free tier with 3 concurrent sessions, no credit card required
  • Live browser preview URL to watch tasks execute in real-time
  • Recording feature produces MP4 video of browser sessions
  • Deterministic rerun with cached scripts up to 99% cheaper and 5-10s fast
  • Structured output with automatic Pydantic/Zod validation and typing
  • Human-in-the-loop support for payment info and complex auth flows
  • SDK available for both Python and TypeScript with async support
  • Follow-up tasks reuse same browser session with cookies and state preserved
  • Workspace file storage for uploading inputs and downloading agent outputs
  • Auto-healing for cached scripts that breaks when website layout changes
  • Streaming live messages showing agent reasoning, tool calls, and browser actions
  • Bring Your Own Key supported for Dev plan and above
  • Scheduled jobs for recurring automation tasks
  • Works with Playwright, Puppeteer via CDP WebSocket connection

Browser Use cons

  • Free tier has no monthly credits, only free browsers with 3 concurrent sessions
  • Maximum session duration limited to 4 hours
  • Sessions timeout after 15 minutes of inactivity by default
  • Recording URLs expire after 1 hour requiring prompt download
  • V3 Agent pricing at 1.2x provider rates plus 0.2x orchestration fee on top
  • Browser session billing at $0.06/hour plus $5/GB proxy bandwidth adds up
  • Custom proxy requires custom plan, not available on lower tiers
  • Auto-healing limited to 1 attempt per run preventing runaway costs but may fail
  • Zod v4 required for TypeScript, v3 not compatible causing migration issues
  • Enterprise features like SLAs and data retention require contact sales

Frequently asked questions about Browser Use

What is Browser Use and what does it do?

Browser Use is the way AI uses the web, providing state-of-the-art AI browser automation. It converts website interfaces into structured text that language models can process faster and more precisely than vision-based approaches that try to see websites like humans. The platform enables AI agents to automate the web with stealth browsers, CAPTCHA solving, residential proxies across 195+ countries, and managed infrastructure. You can use it for data extraction, form filling, multi-step workflows, research, monitoring, and testing websites end-to-end with natural language instructions.

How do I get started with Browser Use?

Get started by installing the SDK: run 'pip install browser-use-sdk' for Python or 'npm install browser-use-sdk' for TypeScript. Then get an API key at cloud.browser-use.com/settings?tab=api-keys and set it as BROWSER_USE_API_KEY environment variable. Your first task can be run with client.run('List the top 20 posts on Hacker News today with their points'). Try for free with no credit card required at cloud.browser-use.com. AI agents can also create a free account autonomously via a challenge-response flow at api.browser-use.com.

What models does Browser Use support?

Browser Use supports Claude Sonnet 4.6 (claude-sonnet-4.6), Claude Opus 4.6 (claude-opus-4.6), and GPT-5.4 Mini (gpt-5.4-mini). The platform recommends Claude Sonnet 4.6 as the model they optimize for the most. Token-based pricing is at 1.2x provider rates: GPT-5.4 Mini costs $0.90 input/$5.40 output per 1M tokens, Claude Sonnet 4.6 costs $3.60/$18.00, and Claude Opus 4.6/4.7 costs $6.00/$30.00 per 1M tokens. You can pass the model parameter to client.run() to select a specific model.

How do stealth browsers work?

Every cloud browser session runs in a hardened Chromium fork with stealth enabled by default and no configuration needed. Anti-detect browser fingerprinting randomizes Canvas, WebGL, fonts, navigator, and other fingerprints per session to appear as a real user, passing CreepJS, BrowserLeaks, and other fingerprint detectors. Ad and cookie banner blocking dismisses banners automatically so agents see clean pages and execute faster. Cloudflare and anti-bot bypass works on sites protected by Cloudflare, PerimeterX, and other bot detection services. Residential proxies are enabled by default across 195+ countries.

What is deterministic rerun with cached scripts?

Deterministic rerun lets you run a browser task once with a full agent (~$0.10, ~60s), then re-execute the same task instantly using a cached script with $0 LLM cost and ~5s execution time, up to 99% cheaper. Use @{{double brackets}} around values that can change between runs. The system strips values to create a template, hashes it to a unique ID, and checks for a cached Python script. If no script exists, the full agent runs and saves a standalone script after completing the task. Future calls with the same template run the script directly with no agent or LLM. Auto-healing detects when scripts break from website changes and automatically regenerates them.

How do proxies work and can I disable them?

A US residential proxy is active by default on every browser session. To route through a different country, set proxy_country_code parameter (e.g., 'de' for Germany, 'jp' for Japan) with 195+ countries supported. Residential proxies are enabled by default and make browser sessions appear as real users from the target geography. If your use case does not need proxies, such as QA testing localhost, you can disable them by setting proxy_country_code to None/null. Custom proxy servers (HTTP or SOCKS5) are available on custom plans with host, port, username, and password configuration.

What is Browser Use Box (BUX)?

Browser Use Box (BUX) is your 24/7 Claude agent with Claude Code and Browser Harness pre-installed in a dedicated remote box. You can drive it from Telegram, the web, or SSH. It is billed hourly while running with three tiers: small at $1/day with 2 vCPU, 4GB RAM, 20GB storage; medium at $2/day with 2 vCPU, 8GB RAM, 50GB storage; and large at $4/day with 4 vCPU, 16GB RAM, 100GB storage. This provides a personal agent that runs continuously rather than on-demand sessions.

How do I get structured output from agents?

Pass a Pydantic model (Python) or Zod schema (TypeScript) to client.run() with output_schema or schema parameter. The result.output is automatically validated and converted to the typed object. For TypeScript, you need Zod v4 (npm install zod@4) as v3 is not compatible. For example, define a Post model with name, points, comments fields and a HNPosts model with posts array, then the agent returns validated typed objects instead of raw text. This ensures reliable parsing of extracted data.

What is the difference between Agent and Browser modes?

Agent mode uses sessions.create() or run() method and runs an AI agent to complete your task with model selection, task execution, and full agent intelligence. Browser mode uses browsers.create() method and provides raw browser access via CDP for manual control without AI. Agent mode supports task, model, keep_alive, workspace_id parameters while Browser mode supports screen size, timeout parameters. Agent mode is for automating tasks with AI while Browser mode is for connecting Playwright/Puppeteer directly to the cloud browser.

How do I watch and record browser sessions?

When creating a session, you get a live_url that shows a live preview of the browser executing your task. You can embed this in your app using an iframe with src set to the live_url. For recording, pass enable_recording=True to client.run() and then call client.sessions.wait_for_recording(result.id) which waits up to 15s for the MP4 recording to be ready. Recording URLs are presigned download URLs for MP4 files that expire after 1 hour, so download promptly. Recording only works when the agent actually opens a browser; tasks answered without browsing produce no recording.

Categories

Use cases

Browse all AI tools on NeedAnAI