Skip to main content

Additional features: Browser use

When to use​

Turn on Browser use when the agent must operate a real web UI—e.g., complete forms in an internal portal, update a CRM record, check ticket status, download a statement, or follow a multi-step workflow that only exists in a browser.

What it does​

  • The agent delegates to a browser agent that opens a live browser session.
  • Interactive elements on each page are highlighted; the browser agent’s LVM decides what to click, which field to fill, and with which values, step by step.
  • You can watch the pages it visits in a side panel as the task runs.
  • The final result (and intermediate steps) are returned to the parent agent, which then replies to the user.

Enable in the UI​

  1. Open your agent → click Edit.
  2. In Additional features, click Add.
  3. Toggle Browser use ON.
  4. Close the drawer; Browser use will appear under Additional features.

How it works (runtime)​

  1. The parent agent decides to browse and sends a task (goal + constraints) to the browser agent.
  2. The browser agent navigates, identifies clickable/typable elements, performs actions, waits for page updates, and repeats until success criteria are met or it times out.
  3. Screenshots/steps appear in the side panel; the outcome returns to the parent for a user-facing summary.

In the Base instruction (parent agent)​

  • Specify allowed domains/URLs, success criteria, and when to ask for confirmation.

Example:

“Use Browser use only for intranet.simplai.ai and jira.simplai.ai. Before submitting any form, show a summary and ask for confirmation. Return the final confirmation number and a screenshot link.”

When asking the agent in chat​

  • Provide: target URL or domain, exact values to fill, selectors/labels if known, and what to return.
  • Examples:
    • “Open helpdesk.simplai.ai/tickets/new, create a P1 ticket: title VPN login failure, user [email protected], team IT. Submit, then return the ticket ID and a screenshot of the success page.”
    • “Go to Jira, search ABC-1421, add the comment ‘Ready for QA’, and move status to In Review. Confirm with the new status and permalink.”

Tip: If login is required, provide credentials via a secure secret/integration (not plain text). Tell the agent which method to use and to stop/ask if MFA or CAPTCHA appears.

Best practices​

  • Constrain navigation: name allowed domains to avoid drift and ads.
  • Be explicit: list form fields and expected values; specify what counts as “done.”
  • Plan → confirm → execute: ask the agent to outline steps, then proceed after confirmation for write actions.
  • Evidence: request a screenshot link, URL, and extracted confirmation number in the final answer.
  • Timeouts & limits: if flows are long, raise per-tool timeout and global tool-call caps as needed (see Tools gear settings).

Known limits & troubleshooting​

  • Auth walls (MFA/CAPTCHAs): the agent will pause/ask; provide one-time codes or complete manually, then resume.
  • Element not found / wrong click: give clearer labels, unique text near the element, or a direct URL; slow pages may need longer timeouts.
  • Downloads: prefer workflows that yield an on-page confirmation or link; file handling may require a dedicated tool.
  • Nothing happens: ensure Browser use is enabled; check tool caps/timeouts; restrict to the correct domain to reduce confusion.

With Browser use enabled, your agent can execute real web workflows end-to-end while you observe every step—bringing UI-only processes into your automated runs safely and transparently.