Browser Use Heroku

--- title: "Sensitive Data" description: "Handle sensitive information securely and avoid sending PII & passwords to the LLM." icon: "shield" --- ## Handling Sensitive Data When working with sensitive information like passwords or PII, you can use the `Agent(sensitive_data=...)` parameter to provide sensitive strings that the model can use in actions without ever seeing directly. ```python agent = Agent( task='Log into example.com as user x_username with password x_password', sensitive_data={ 'https://example.com': { 'x_username': 'abc@example.com', 'x_password': 'abc123456', # 'x_placeholder': '<actual secret value>', }, }, ) ``` <Note> You should also configure [`BrowserSession(allowed_domains=...)`](https://docs.browser-use.com/customize/browser-settings#allowed-domains) to prevent the Agent from visiting URLs not needed for the task. </Note> ### Basic Usage Here's a basic example of how to use sensitive data: ```python from dotenv import load_dotenv load_dotenv() from browser_use.llm import ChatOpenAI from browser_use import Agent, BrowserSession llm = ChatOpenAI(model='gpt-4.1') # Define sensitive data # The LLM will only see placeholder names (x_member_number, x_passphrase), never the actual values sensitive_data = { 'https://*.example.com': { 'x_member_number': '123235325', 'x_passphrase': 'abcwe234', }, } # Use the placeholder names in your task description task = """ 1. go to https://travel.example.com 2. sign in with your member number x_member_number and private access code x_passphrase 3. extract today's list of travel deals as JSON """ # Recommended: Limit the domains available for the entire browser so the Agent can't be tricked into visiting untrusted URLs browser_session = BrowserSession(allowed_domains=['https://*.example.com']) agent = Agent( task=task, llm=llm, sensitive_data=sensitive_data, # Pass the sensitive data to the agent browser_session=browser_session, # Pass the restricted browser_session to limit URLs Agent can visit use_vision=False, # Disable vision or else the LLM might see entered values in screenshots ) async def main(): await agent.run() if __name__ == '__main__': asyncio.run(main()) ``` In this example: 1. The LLM only ever sees the `x_member_number` and `x_passphrase` placeholders in prompts 2. When the model wants to use your password it outputs x_passphrase - and we replace it with the actual value in the DOM 3. When sensitive data appear in the content of the current page, we replace it in the page summary fed to the LLM - so that the model never has it in its state. 4. The browser will be entirely prevented from going to any site not under `https://*.example.com` This approach ensures that sensitive information remains secure while still allowing the agent to perform tasks that require authentication. --- ### Best Practices - Always restrict your sensitive data to only the exact domains that need it, `https://travel.example.com` is better than `*.example.com` - Always restrict [`BrowserSession(allowed_domains=[...])`](https://docs.browser-use.com/customize/browser-settings#allowed-domains) to only the domains the agent needs to visit to accomplish its task. This helps guard against prompt injection attacks, jailbreaks, and LLM mistakes. - Only use `sensitive_data` for strings that can be inputted verbatim as text. The LLM never sees the actual values, so it can't "understand" them, adapt them, or split them up for multiple input fields. For example, you can't ask the Agent to click through a datepicker UI to input the sensitive value `1990-12-31`. For these situations you can implement a [custom function](/customize/custom-functions) the LLM can call that updates the DOM using Python / JS. - Don't use `sensitive_data` for login credentials, it's better to use [`storage_state`](docs.browser-use.com/customize/browser-settings#storage-state) or a [`user_data_dir`](/customize/browser-settings#user-data-dir) to log into the sites the agent needs in advance & reuse the cookies: ```bash # open a browser to log into the sites you need & save the cookies $ playwright open https://accounts.google.com --save-storage auth.json ``` Then use those cookies when the agent runs: ```python agent = Agent(..., browser_session=BrowserSession(storage_state='./auth.json')) ``` <Warning> Warning: Vision models still see the screenshot of the page by default - where the sensitive data might be visible. It's recommended to set `Agent(use_vision=False)` when working with `sensitive_data`. </Warning> <a name="allowed_domains"></a> <a name="domain-pattern-format"></a> ### Allowed Domains Domain patterns in `sensitive_data` follow the same format as [`allowed_domains`](https://docs.browser-use.com/customize/browser-settings#allowed-domains): - `example.com` - Matches only `https://example.com/*` - `*.example.com` - Matches `https://example.com/*` and any subdomain `https://*.example.com/*` - `http*://example.com` - Matches both `http://` and `https://` protocols for `example.com/*` - `chrome-extension://*` - Matches any Chrome extension URL e.g. `chrome-extension://anyextensionid/options.html` > **Security Warning**: For security reasons, certain patterns are explicitly rejected: > > - Wildcards in TLD part (e.g., `example.*`) are **not allowed** (`google.*` would match `google.ninja`, `google.pizza`, etc. which is a bad idea) > - Embedded wildcards (e.g., `g*e.com`) are rejected to prevent overly broad matches > - Multiple wildcards like `*.*.domain` are not supported currently, open an issue if you need this feature The default protocol when no scheme is specified is now `https://` for enhanced security. For convenience the system will validate that all domain patterns used in `Agent(sensitive_data)` are also included in `BrowserSession(allowed_domains)`. ### Missing or Empty Values When working with sensitive data, keep these details in mind: - If a key referenced by the model (`<secret>key_name</secret>`) is missing from your `sensitive_data` dictionary, a warning will be logged but the substitution tag will be preserved. - If you provide an empty value for a key in the `sensitive_data` dictionary, it will be treated the same as a missing key. - The system will always attempt to process all valid substitutions, even if some keys are missing or empty. --- ### Full Example Here's a more complex example demonstrating multiple domains and sensitive data values. ```python from dotenv import load_dotenv load_dotenv() from browser_use.llm import ChatOpenAI from browser_use import Agent, BrowserSession llm = ChatOpenAI(model='gpt-4.1') # Domain-specific sensitive data sensitive_data = { 'https://*.google.com': {'x_email': '...', 'x_pass': '...'}, 'chrome-extension://abcd1243': {'x_api_key': '...'}, 'http*://example.com': {'x_authcode': '123123'} } # Set browser session with allowed domains that match all domain patterns in sensitive_data browser_session = BrowserSession( allowed_domains=[ 'https://*.google.com', 'chrome-extension://abcd', 'http://example.com', # Explicitly include http:// if needed 'https://example.com' # By default, only https:// is matched ] ) # Pass the sensitive data to the agent agent = Agent( task="Log into Google, then check my account information", llm=llm, sensitive_data=sensitive_data, browser_session=browser_session, use_vision=False, ) async def main(): await agent.run() if __name__ == '__main__': asyncio.run(main()) ``` With this approach: 1. The Google credentials (`x_email` and `x_pass`) will only be used on Google domains (any subdomain, https only) 2. The API key (`x_api_key`) will only be used on pages served by the specific Chrome extension `abcd1243` 3. The auth code (`x_authcode`) will only be used on `http://example.com/*` or `https://example.com/*`