Skip to main content

​
Task is slow

  • Switch models. Browser Use LLM is ~3s/step. Frontier models (Claude, Gemini, O3) are 6-8s/step. See Models & Pricing.
  • Set a start URL. Skip navigation steps: start_url="https://example.com/page".
  • Use a closer proxy. Proxy distance adds latency. If your target site is in the US, use proxy_country_code="us".

​
Agent failed or gave wrong output

  • Check the live URL. Create a session manually and watch what the agent does via session.live_url.
  • Simplify instructions. Break complex tasks into smaller steps. Run multiple tasks in the same session.
  • Add a start URL. Don’t make the agent search for a site — send it directly there.

​
Need to log in to a site

Three options, from easiest to most secure:
  1. Profile Sync — sync your local browser cookies. One command: curl -fsSL https://browser-use.com/profile.sh | sh. See Profiles.
  2. Secrets — pass credentials scoped by domain. See Authentication.
  3. 1Password — auto-fill passwords and 2FA/TOTP codes. See Authentication.

​
Getting blocked by a website

Stealth is on by default (anti-fingerprinting, CAPTCHA solving, Cloudflare bypass). If you’re still blocked:
  • Try a different proxy country. Some sites geo-restrict. Use proxy_country_code to match the target region.
  • Slow down. Use flash_mode=False for more careful navigation.
  • Check if the site uses a login wall. Use a Profile to authenticate first.

​
How much does a task cost?

$0.01 init + per-step cost based on model. A typical 10-step task with Browser Use LLM costs $0.03. See Pricing for full details.

​
Rate limited (429 errors)

The SDK auto-retries 429 responses with exponential backoff (up to 10s). If you’re consistently hitting limits, you may need to upgrade your plan for more concurrent sessions. See Pricing.