Employee browser
Watch an employee browse, sign in yourself, and take over the same page.
Open an employee's channel. The Browser tab appears in the right pane. Choose Enable browser to create your personal session for that employee.
Enter a website address and sign in yourself. Passwords, verification codes and website challenges belong in the browser pane. Keep the session in your control until sign-in finishes, then choose Hand back and give the employee a clear task in chat.
For example: “Find five founders working in the AI space. Follow exactly five eligible profiles, skip anyone already followed, and report the five profiles whose buttons changed to Following. Do not send messages or connection requests.”
The employee uses the same page you see. Choose Take over whenever you need to intervene. This stops the employee from reading or changing the page until you hand it back. Website dialogs offer OK, Cancel and a response field when needed.
Follow the work in chat
Chat shows a numbered Steps trail as the subscription employee completes browser actions: opening a website, inspecting page elements, reading visible text, clicking, scrolling, and filling a form. Recent steps stay visible while it works and underneath the completed reply. Choose Show all steps to expand the retained history.
Steps describe completed tool actions, including failures. They appear after each action returns and the employee starts its next model request, followed by the normal chat refresh. The trail retains up to 60 recent actions. It does not copy field values, page contents or URL query strings. Element observations and visible page text are how the employee reads the site; an explicit image is available when text is insufficient.
For research, the final reply should cite visited sources, explain which observed facts support its conclusions, and identify inferences or missing evidence. Use the trail to follow the browsing and the source links to assess the findings.
Task scope and permissions
A browser session is personal to its owner and assigned employee. Another workspace member cannot use your login merely because they administer the company. Supervised employees request approval for generic browser changes because a click can submit a message or complete a purchase. Explicitly selected full autonomy follows the employee's existing permission mode.
Browser tools currently integrate with the supported subscription employee runtime. If your employee's provider path does not expose browser tools, use a supported employee for the workflow; do not paste browser or model credentials into its chat.
Session settings
The gear menu contains Stop, Reopen, Import cookies and Delete. Normal stops and hand-back save encrypted login state. A browser may still ask you to sign in again after a challenge, device-bound check or unexpected crash. Cookie import is optional and does not replace a manual sign-in for unsupported sites.
The live pane keeps no screenshot history. An employee can take an inline screenshot when visual evidence is useful. Saving screenshots and recording the browser are currently unavailable, so browser captures cannot be attached from a saved workspace file. Treat any page content included in an employee task as information shared with its configured model provider.
Keep recordings of real signed-in accounts private. Share product demonstrations using mock accounts and synthetic data.
Approvals and autonomy
How Yodu's human-in-the-loop model works — what triggers an approval, the inline chat card and the queue, Allow once / This session / Deny, and how the gate composes with per-employee tool switches.
Schedules and recurring work
Create, test, pause, edit, and review recurring AI employee jobs in Yodu with clear owners, timing, run history, and approval controls.