Browser start and stop

An agent can start its assigned browser when an authorized task needs it, in either conversation approval mode. The viewer can stay closed. The agent continues its original browser request when startup finishes.

The agent can stop the browser and start it again. An admin must still approve the agent’s browser access and permitted websites. Saved sign-ins remain in the same browser profile. After a restart, the agent opens the required page and checks whether sign-in is still active.

The same browser access applies to authorized email, Slack, Teams, Telegram, WhatsApp, and Discord tasks. Browser startup does not approve new websites or other tools. Retained work checks its original sender and destination access before continuing.

Private human control, deliberate user pauses, profile grants, and website restrictions still apply. An agent cannot take over a private session or use a browser held by another conversation. Return control through the existing browser controls when you are ready.

Continue after a connection interruption

A connection interruption does not mean that you paused the agent. After an automatic browser shutdown, authorized work can continue with the same saved browser. You do not need to open the viewer. The agent checks the page before it repeats an action whose result is unknown.

A deliberate pause or private sign-in still requires you to return control. A website can expire its own sign-in even when the saved browser remains available.

Use linked website credentials

An admin can link a saved credential to a project for one exact HTTPS website origin. Browser-only links do not make credentials available to coding commands. The agent uses the link name to fill a field; it does not need the credential in the conversation. Filling a field does not submit the form. After credential use, the agent reads text snapshots. Agent screenshots are unavailable until the browser is stopped and started again.

If a credential expires, is revoked, or is no longer linked, the agent stops and asks for the link to be checked. Private control and the agent's website permissions still apply. Use private sign-in for additional verification such as a security code or CAPTCHA.

Browser credential links accept HTTPS origins on the standard port 443. Protected filling requires values with at least four characters. Refill a shortened credential field before asking the agent to read the page. Browser links remain visible in agent settings while command credential access is off. Remove link removes the selected browser link.

Credential values are stored in Settings → Secrets & variables. Select the workspace under Settings scope. In the agent’s settings, Secrets & variables shows its linked browser credentials and permitted website. Creating a website-specific browser link currently requires the API; the app can show and remove that link.

Idle shutdown and capacity

Without an open viewer, a browser stops after ten minutes without browser actions. A new browser action resets this timer. An open viewer keeps the browser running. Saved sign-ins remain in the profile, although the website can expire its own session.

Browser capacity is shared within a deployment. If capacity is full, stop another browser and retry. Each saved browser keeps its agent grants and sign-ins.