Recording

The recorder turns what you do in the browser and in Windows programs into commands: click Record, do the work once, click Stop, and the steps appear in your task, with every element captured as if you had used the Capture button.

#Before you start

• The Botream Agent tray app must run on the computer you are working on.
• To record in the browser, the Botream Element Capture extension must be installed there (see Install the extension).
• Desktop recording runs only in the tray app, and not while the agent is running a task.
• When several agents are online, the Record dialog asks you to choose the one on this computer, then Start recording.

#Recording a task

1. In the editor, select the block after which the steps should go. If that block uses a browser page or a window variable, the recording uses it too (see How steps become commands).
2. Click Record in the editor toolbar. The Record dialog shows "Recording: do the steps in Chrome or in your Windows programs".
3. Switch to the browser or program and do the work. The editor stays where it is.
4. Click Stop in the red bar at the bottom of the page or of the screen, or Stop and add the steps in the editor. Discard throws the recording away.

A recording ends by itself after 30 minutes.

While you record:

• A dashed red frame shows the element under the mouse.
• Each recorded step flashes green with what was noted.
• The bar at the bottom shows the number of steps, the last step and Stop.

#What is recorded in the browser

Keys that only edit a field (typing, Backspace, arrows, Ctrl+A/C/V/X/Z/Y) are left to the field's settext. Ctrl+wheel zoom is not recorded.

#What is recorded on the desktop

The desktop recorder watches the mouse and keyboard without getting in their way; the program receives every click and key as usual.

Keys are recorded with the keyboard layout of the window they go to, including AltGr characters such as @ and [.

Not recorded on the desktop: the browsers the extension already records, the editor window you started from, and the Botream Agent itself.

#Passwords

Passwords are never recorded:

• A browser password field is recorded with an empty value.
• On the desktop, keys typed into a password field are left out.

In both cases the recording adds a comment in their place: "Enter the password here: keep it in a vault and use a credential variable ($cred.Password$)". Replace it with a credential variable.

#How steps become commands

The steps are ordered by time and inserted after the selected block.

#Browser steps

• The first page becomes browser --action:open --url:… --output:page. The variable is named page, or page2, page3… when that name is taken.
• When the selected block already uses a browser (--browser:$x$, or a browser --action:open --output:x), the recording uses that page instead and adds no open.
• Each step becomes capture --browser:$page$ --action:… with --dompath and the Object properties, exactly as a capture would give them. Keys become capture --action:sendkeys --keys:….
• Addresses typed in become browser --action:navigate.
• Wheel turns become browser --action:scroll --amount for the page, or capture --action:scroll --dompath --amount for a panel.

#Desktop steps

• The first time a window is used, the recording adds window --action:attach --path:<program> (store apps: --title) with a new window variable, unless it can reuse one: the selected block's window variable when its program or title matches, or a window variable the task already has for that program.
• Each click becomes a desktop --window:… --action:… --by --target --properties command.
• Typing becomes sendkeys --window:… --keys:….
• Wheel turns and drags become mouse --window:… --action:scroll and mouse --window:… --action:drag, in pixels from the window's corner: they follow the window when it moves, but not changes to its layout.

browser --action:open --url:"https://shop.example.com/login" --output:page
capture --browser:$page$ --action:settext --dompath:"//input[@id='email']" --value:"anna@example.com"
comment --text:"Enter the password here: keep it in a vault and use a credential variable ($cred.Password$)"
capture --browser:$page$ --action:settext --dompath:"//input[@id='password']" --value:""
capture --browser:$page$ --action:click --dompath:"//button[normalize-space(.)='Sign in']"
window --action:attach --path:"erp.exe" --output:window1
desktop --window:$window1$ --action:click --by:automationid --target:"btnNewOrder"
sendkeys --window:$window1$ --keys:"10045[TAB]"

The result is ordinary commands: edit them as usual. Fallback locators, anchors and self-healing work for recorded elements, because they carry their object properties.

#Recording + AI

When you stop, the dialog shows what was recorded (for example "Recorded: 12 steps in 2 pages") and two choices:

• Insert as recorded adds the commands as they are.
• Improve with AI lets the AI model turn them into a clean task first.

#What the AI changes

• Removes noise: double clicks, focus clicks before typing, needless scrolling.
• Groups the commands into named steps.
• Turns typed values into task inputs (the recorded value becomes the default), and repeated steps into a loop over a list input.
• Adds waits after page changes.
• Replaces password notes with a credential variable.

#Elements never change

The AI may reuse, reorder or drop the recorded elements, but it may never invent or edit one: every selector, desktop locator and object properties table in the result must be one that was recorded. A result that breaks this rule, or is not valid, goes back to the model once. If it still has problems, they are listed under "The refined version cannot be used as it is", and only the recording can be inserted.

#Comparing and choosing

The dialog shows both versions side by side, As recorded and Refined by the model's name, with the model's list of changes below. Then:

• Use the AI version inserts the refined task after the selection; its input and output declarations move to the top of the task.
• Insert as recorded inserts the original commands.
• Refine again asks the model for another version.

Note: Improve with AI needs an AI model configured by an administrator, and the Edit tasks permission (Developer by default). See AI features.

Tip: The AI Task Generator can also explore a site in a browser by itself and capture the elements as it goes; its result goes through the same refinement. See AI features.