Write scripts without coding
What a script is, adding and editing steps, picking XPath with the 🎯 button, using Excel columns and variables, If / Repeat blocks, shortcuts, the Market and the full command reference.
A script is the list of steps the tool performs for each account, saved as a .json file. A login script, for example: Open page → Wait for the Email box → Type {@Email} → Type {@Pass} → Click Log in → Wait for the home page.
The Script page: 3 columns
- Command library (left): search commands in Vietnamese with or without accents, or in English. Click a command to add it to the script.
- Step list (middle): indented by If / Repeat blocks. Drag the step number to move a step.
- Editor (right): click a step and the right panel shows its fields. Rarely used fields sit under ⚙️ Advanced.
With no step selected, the right panel offers: Save, Save as…, New script, Record actions, Capture sample image, Test the whole script and the shortcut table.
Showing the tool where to click (XPath)
You don't need to write XPath. Next to the Target box there is a 🎯 button: press it, then click the button on the open Chrome page and the XPath fills in by itself.
|, e.g. Skip for now|Skip|Not now.Excel data and variables
| Write | Meaning | Example |
|---|---|---|
{@ColumnName} | The Excel value of the running account | {@Email}, {@Pass} |
{$name} | A variable the tool picked up during the run | {$code} after reading an email code |
{name}, {stt} | Profile name, account number | Naming report files |
If your Excel has a Proxy column, each account takes the proxy on its own row.
If / Repeat / Branches
- If… (element exists, title contains text, variable equals…) → the commands below only run when true. You can add Else, and it must be closed with End if.
- Repeat N times / Repeat for N seconds → close with End repeat. Leave early with Break loop.
- Skip this account: use inside an If block to handle an account blocked midway (write the report, close the profile, continue with other accounts).
- Stop (success): finish the account right away and count it as done, e.g. the account already did today's task.
Working fast with many steps
Select several steps (click / Ctrl / Shift / drag) to run, duplicate, delete, put them into a 🏷 Group or add a 📝 Note. Right-click opens a menu like in any normal app.
| Key | Action |
|---|---|
| Ctrl+A / C / X / V | Select all / copy / cut / paste steps |
| Ctrl+D | Duplicate step |
| Delete | Delete step |
| Alt+↑ / Alt+↓ | Move step up / down |
| Enter or F2 | Edit step |
| Ctrl+Enter | Test the selected step |
Recording and the Market
- Record actions: press record, do the job by hand in the browser and the tool copies it into steps. Then let the AI clean it up and turn typed text into Excel columns (AI assistant).
- Market: sample scripts by group. Insert into script (after the selected step) or Use as a new script. Save your own scripts into the
kich_ban_maufolder and they appear under Yours.
Full command reference
The list below is read from the running tool (version 3.0+ffix-9p17C2+x2.2+xl1.1+direct-7+cap1.0), including add-on commands (X, wallets, AI…). Newly added commands carry a NEW label.
Wait · 5 commands
waitNEWPause a few seconds. Type 3 or 2-5 (random).
More detail
Pause before the next command. For pages that need time to load. Tip: avoid a fixed number for every account (looks like a bot) — use 2-5 seconds varied, or better, 'Wait for element' to wait until the button really appears.
wait_forNEWWait until the element appears / disappears.
More detail
Instead of blindly waiting a few seconds, this waits until AN ELEMENT appears (or disappears). E.g. after Login, wait for 'Home' before continuing. Pick the element with 🎯 'Pick' then click it on the page. Times out with an error (retry possible).
wait_urlNEWWait until the current URL contains this text.
More detail
Wait until the tab's address bar contains some text. Great after pressing Login: wait for 'home' or 'dashboard' in the URL to know you are in. Only PART of the URL is needed. Times out with an error.
wait_titleNEWWait until the tab title contains this text.
More detail
Like 'Wait for URL' but watches the tab TITLE. Use when the URL stays the same but the title changes, e.g. 'Login' → 'Home'.
wait_tabNEWWait until a new tab appears (e.g. after opening a wallet popup) then switch to it.
More detail
Many buttons open a NEW TAB (wallet window, Google login…). This waits for it and switches so later commands work in the right tab.
• 'A new tab appears': tab count grew since the command started.
• 'At least N tabs': wait for N tabs.
• 'At most N tabs left': wait for popups to close.
Times out with an error if no tab appears.
Interaction · 16 commands
clickNEWClick an element / button.
More detail
Click a button/element on the page. Pick the element: press 🎯 'Pick', then click it on the account's page → the tool fills in how to find it (xpath). If several identical elements exist, use 'Order' to choose first / last / Nth / random. For a 'human-like' click (curved mouse path) use the DIRECT group → 'Click element (human-like)'.
click_textNEWClick a button by the TEXT on it (e.g. Skip for now, Next) — no XPath/CSS, independent of window size. Several options separated by | (e.g. Skip for now|Skip|Not now). Tick 'ignore errors' when the button does not always appear.
More detail
Easiest for beginners: NO element picking, just type the button text. Button says 'Skip for now' → enter Skip for now. Several options separated by |: Skip for now|Skip|Not now (tried in turn). Match 'exact' = whole phrase, 'contains' = just contains the text.
dblclickNEWDouble click the element.
More detail
Double click (2 quick clicks) an element. For places that only open on double click, e.g. opening a file, selecting a word. Pick the element with 🎯 Pick like the Click command.
hoverNEWHover the element (shows menu...).
More detail
Hover an element without clicking — for menus that only appear on hover.
click_xyNEWClick at point x, y on the page (px from top-left).
More detail
Click a point by x, y coordinates instead of picking an element. x = px from the LEFT edge, y = px from the TOP edge. Use when the target has no clear element (e.g. drawn on a canvas). To find coordinates use 'Get element position & size' or 'Find image and get coordinates'.
move_xyNEWMove the mouse to point x, y.
More detail
Move the mouse to point x, y (no click). To hover a menu so it opens, or for human-like behavior. x from the left edge, y from the top edge.
fillNEWFast fill. Use {1} {2}... for data, {$variable}, {@ExcelColumn}.
More detail
Quickly fills an input (name, password…). Content can mix: {1} = column 1 of pasted data, {@Username} = the Username column in Excel, {$variable} = from an earlier command. E.g. {@Email} or user{stt}@mail.com. For a more human feel use 'Type key by key' or 'Send text' with 'Real keyboard'.
typeNEWType slowly character by character like a human.
More detail
Types slowly key by key like a human, unlike 'Fast fill' (one paste). Use it on sites that check typing speed. Content may use {1} {$variable} {@Column} like Fill.
send_textNEWSend text into a field from many sources: manual, Excel column, text file, pasted list (one line per account, in order or random) or auto-generated (username, password, email, name...).
More detail
The 'all-purpose' command to put text into a field, from many sources:
• Manual: type the content yourself.
• Excel column: each account takes the text from its own row (e.g. Password column).
• Text file: one value per line, each account takes one (in order / random / unique / remembered cursor).
• Auto random: username, password, full name, email, phone number…
The value is saved to a variable (default {$user}) so a later command can write it to file. 'Send mode' = Fast fill / Key by key / Real keyboard (typos can be enabled for realism).
type_focusNEWType into the focused field, no element needed.
More detail
Type into the FOCUSED field (blinking cursor), no element needed. Use right after clicking a field. Keys are typed with delays.
pressNEWPress key: Enter, Tab, Escape, Control+A...
More detail
Press a key: Enter, Tab, Escape, ArrowDown… or a combo like Control+A (select all), Control+C. E.g. press Enter after typing in a search box. Use the exact capitalization: Enter, Tab, Escape, Backspace, ArrowLeft.
selectNEWSelect an item in a dropdown. Can pick RANDOMLY.
More detail
Select an item in a dropdown ('Country', 'Birth month'…). Select by visible text, hidden value, index, or RANDOMLY within a range you set (good for registering many accounts, each different).
checkNEWMake sure the box is ticked.
More detail
Make sure a checkbox IS ticked. Already ticked stays ticked, so running twice is safe.
upload_imgNEWPick one image (random) from a folder on the PC and load it into the upload field. Target the <input type=file> element.
More detail
Load an image into an upload field (e.g. change avatar). Pick a folder on the PC; each account gets one image (random / unique / in order). The target is the <input type=file> element (often hidden) — if 🎯 cannot grab it, use 'Upload file (exact file)' or an xpath containing input type=file.
uncheckNEWMake sure the box is unticked.
More detail
Make sure a checkbox is NOT ticked. Already unticked stays unticked, so running twice is safe.
scroll_toNEWScroll until the element is visible.
More detail
Scroll until an element is visible. For buttons far down the page that cannot be clicked until scrolled into view.
Get value · 16 commands
mail_codeNEWReads the verification code from the account's Hotmail/Outlook (email|pass|refresh|client_id|recovery) and fills it in. Saved to {$code}.
More detail
Reads the verification code from the account's Hotmail/Outlook inbox and fills it in. Needs the account as email|pass|refresh_token|client_id|recovery. Set the code length (usually 6) and a keyword in the mail (e.g. 'code') to avoid picking the wrong mail. Code saved to {$code}.
mail_checkNEWReads the account's Hotmail inbox for keywords (e.g. suspended, locked…). Match saved to {$mail_hit} (empty = none, 'loi' = unreadable). Can auto-skip the account on match.
More detail
Reads the account's Hotmail inbox (like Get Hotmail Code) and searches keywords, e.g. 'suspended, locked' to filter locked accounts. A match is saved to {$mail_hit} (no match = empty, read error = 'loi'); choose 'Skip this account' to skip it and continue, optionally writing skipped accounts to CSV. For bulk viewing / filtering outside scripts press 📧 Hotmail at the top.
get_textNEWRead the text (or input value) into a variable.
More detail
Reads an element's text (or input value) into a variable. E.g. wallet balance → {$sodu}; later use {$sodu} to compare or write to file.
get_attrNEWRead attributes like href, src, value, class...
More detail
Read a 'hidden attribute' of an element, like href (link), src (image path), value, class. E.g. the ref link of a Copy button: attribute href.
get_countNEWCount how many elements match.
More detail
Count how many elements match your selector and save the number to a variable. E.g. count visible 'Follow' buttons to know how many clicks.
get_urlNEWSave the current tab's URL to a variable.
More detail
Save the current tab's URL to a variable. E.g. to check whether you were kicked to the login page, or to read a code from the URL.
get_titleNEWSave current tab title to a variable.
More detail
Save the current tab's title to a variable.
get_jsNEWRun JavaScript and save the result to a variable (for those who know JS).
More detail
FOR THOSE WHO KNOW JAVASCRIPT. Runs a small snippet on the page and saves the result to a variable. E.g. document.title or window.scrollY . Non-coders should use the other 'Get value' commands.
tele_infoNEWRead the name, username, phone of the logged-in Telegram Web account and write to csv.
More detail
WHEN: the profile is logged into Telegram Web and you want its name / username / phone.
HOW: the script needs only this one command (the tool opens web.telegram.org itself if elsewhere).
The tool ONLY READS Telegram Web's local storage, clicks nothing, sends nothing.
RESULT: one csv line (Profile name, Telegram name, Username, Phone, User ID, Status, Time) + variables {$tg_ten} {$tg_username} {$tg_sdt} {$tg_id} {$tg_trangthai} for later.
A logged-out account still writes one 'Not logged in' line, not counted as an error.
otpNEWGenerate a 6-digit code from the 2FA secret (Base32) for the code field.
More detail
Generate a 6-digit 2FA code (like Google Authenticator) from the secret key. The secret is the long Base32 string from when 2FA was enabled, usually in an Excel column: enter {@2FAColumn}. Codes change every 30 seconds, so place this right before filling the code.
get_focusNEWSee which element has focus (being typed in): tag, id, name, value… saved to variables.
More detail
Focus = the field currently selected for typing (blinking cursor). Checks 'did I click the right field' before typing.
E.g. get the field's 'name' → then If value: if {$o} = email type the email, if = password type the password.
If nothing is focused the variable is empty.
get_boundNEWGet an element's top-left, width, height and center into {$name_x} {$name_y} {$name_w} {$name_h} {$name_cx} {$name_cy}.
More detail
Useful for coordinate-based DIRECT commands: get the element center (cx, cy) then 'Click coordinates' x = {$ten_cx}, y = {$ten_cy}.
Coordinates in px from the viewport's top-left. Off-screen elements are scrolled to the middle before measuring.
mail_harvestNEWRight after logging into Hotmail in the browser (after B10), grant OAuth2 and save a NEW refresh_token to file — so later runs read mail via mail_code/mail_link without a browser login.
mail_linkNEWReads Hotmail with the existing refresh_token (NO browser login), extracting a LINK from the mail (e.g. Reset Password) instead of just a code.
mail_harvest_batchNEWFor a shared recovery-mail pool (e.g. 6 mails): paste an email|pass list (one per line); the action logs into each mail in turn, grants OAuth2, writes the new refresh_token to one shared file. Failed mails are skipped. ONE SINGLE STEP — no need to repeat B4/B9/B10 per mail. Chrome/Edge only.
mail_pickNEWClicks/reads NOTHING on the page — only takes the EMAIL ADDRESS (not a code) of one line in the account list, with the same rotation as mail_code, and saves it to a variable. Leave 'Select element by / XPath' above blank. Use the variable in a separate send_text step to fill 'Add an email address'.
Navigation · 11 commands
gotoNEWOpen a URL in the current tab.
More detail
Open a URL in the current tab. Include https://. For heavy/slow pages add 'Wait for page load' or use 'Open URL — wait full load' instead.
new_tabNEWOpen a new tab, optionally with a URL.
More detail
Open a new tab (optionally with a URL) and switch to it. To go back use 'Switch window / tab'.
switch_tabNEWSwitch to a tab / popup (e.g. wallet popup). 0 = first, 1 = second..., -1 = last.
More detail
When the site opens a popup/tab (wallet popup, Google login), the tool still works in the old tab. This switches tabs: 0 = first, 1 = second, -1 = newest. Not sure which → use 'Switch tab by text'.
close_tabNEWClose the current tab, switch to the last remaining one.
More detail
Close the current tab, the tool switches to a remaining one.
close_other_tabsNEWKeep current tab, close the rest.
More detail
Keep the current tab, close all others. Cleans up ad / popup tabs.
backNEWGo back one page.
More detail
Go back one page (browser's left arrow).
reloadNEWReload page (F5).
More detail
Reload the page (like F5). When the page hangs or data needs refreshing.
scrollNEWScroll down / up a bit or to the top / middle / bottom.
More detail
Scroll the page in the background (no real mouse). Down/up by px (500 ≈ half a screen), or jump to top / bottom.
screenshotNEWSave a page screenshot to file. {stt} {name} allowed in the name.
More detail
Save a screenshot of the current page to a file. The name may use {stt} and {name} so each account gets its own file, e.g. anh/{name}.png. Good as proof of results or for reviewing errors.
clear_cookiesNEWClear all cookies of the open browser.
More detail
Clear all cookies (login state) of the running profile. CAREFUL: the account gets logged out everywhere. Only for a clean reset.
scroll_xyNEWScroll the page to exact position x, y (or scroll by x, y more).
More detail
'Scroll to' = move the page to that exact position from the top (y = 0 is the top). 'Scroll more' = scroll from the current position; positive y is down, negative is up.
For human-like scrolling pick 'smooth' (the page glides instead of jumping).
Frame · Tab · Cookie · 18 commands
frame_inNEWEnter a 'child frame' of the page so you can click / type elements inside it.
More detail
WHEN: many sites embed part of the page in a 'child frame' (iframe) — card fields, captcha boxes, wallet popups… 'Element not found' errors are often because it sits in a frame.
HOW: ① this command (enter frame) → ② normal click / type commands → ③ 'Leave frame' when done.
PICKING THE FRAME: easiest is 🎯 Pick then click the frame itself; or 'Frame index' (1 = first); or 'Frame URL contains' (e.g. recaptcha).
TIP: not sure which frame? Run 'List frames' to print them to the Log.
frame_outNEWAfter finishing inside the frame (iframe), return to the main page.
More detail
Use after 'Enter frame'. The tool also returns to the main page whenever you open a new URL / switch tab.
frame_listNEWPrint the page's frames to the Log (index · URL · name).
More detail
Use when you do not know which frame holds the element. Test this command, check the Log: frame 1, 2, 3… then use that number in 'Enter frame'.
goto_waitNEWOpen a URL and CHOOSE the load level: light / full / network idle.
More detail
WHEN: heavy sites (wallets, exchanges, ad-heavy pages) — plain 'Open URL' often moves on before the page is ready.
• Light: like plain 'Open URL', fastest.
• Full: wait for page + images + scripts.
• Very safe: wait until the page STOPS calling the network (safest, a bit slow).
On timeout, if the page already has text the tool continues (with a warning).
wait_loadNEWOpens nothing — just waits for the current page to load to the level you choose (use after a click that navigates).
More detail
WHEN: after a Click that navigates to a new page (e.g. 'Login'), add this to wait for the new page to load before continuing.
switch_tab_byNEWSwitch to the tab whose URL / title contains your text (no need to count tabs).
More detail
WHEN: a click opens a new tab / window (Google login, MetaMask…) — switch to it by text in the URL or title, e.g. 'accounts.google'.
tab_popupNEWSwitch to the NEWEST window / tab that just opened (wallet popup, Google login…). From here every command runs in the popup.
More detail
WHEN: a click opens a popup window (Gate / MetaMask asking for a password, Google login…). Add this right after the click → the tool waits for the popup and switches to it. Every later command (Click, Send text, Wait for element, real mouse…) runs in the popup. When done add 'Return to main window'.
'Only accept window containing': blank = newest; e.g. Gate Wallet = only a window whose title or URL contains that.
Testing commands one by one works too: the tool remembers it is in the popup.
tab_mainNEWReturn to the main tab (the one the script started in). Later commands run there again.
More detail
Use after finishing in the popup (e.g. wallet password entered, Connect pressed). Whether the popup closed itself or not, the tool finds the main tab again.
tabs_listNEWWrite to variable + Log: index | title | URL of every tab.
More detail
Check how many tabs are open and which. One line per tab: 1|Title|https://…
new_tab_privateNEWOpen a tab in an incognito window — WITHOUT the profile's cookies / logins.
More detail
WHEN: you need to view a page as a logged-out stranger. NOTE: Chrome (Chromium) profiles only; Firefox not supported. Incognito tabs do not use the profile's cookies, so not for logged-in work.
cookie_getNEWRead the current page's cookies into a variable — to save or reuse for another account.
More detail
Cookies are the site's 'login memory card'. Get them after logging in and save with 'Append line to txt file': next time 'Load cookies' logs you straight in.
• Blank 'Cookie name' = get all.
• One-line format (a=1; b=2) is easiest to paste back.
cookie_setNEWPaste cookies (one line a=1; b=2, or JSON from an extension) so the browser is already logged in.
More detail
HOW: ① (recommended) open the site first with 'Open URL' ② paste cookies ③ reload. You may use {$cookie} from 'Get cookies' or {@Cookie} from an Excel column.
One-line cookies need to know their site → fill 'Apply to site' (e.g. https://x.com) if that page is not open yet.
get_listNEWGet the text (or link, image…) of ALL similar elements into one variable, one per line.
More detail
EXAMPLE: get every post link on the page → pick the link tags (Tag = a), choose 'Attribute', type href. The result is in {$danhsach}, one link per line — save with 'Append line to file'.
Only counts elements visible on the page.
click_allNEWClick every matching element in turn (e.g. every 'Follow' button, every checkbox) with random pauses.
More detail
EXAMPLE: 20 'Follow' buttons → pick by text 'Follow', 'Click at most' = 5, pause 2-4 seconds. The tool clicks the first 5 with random 2-4 second pauses, human-like.
Only visible elements. Tick 'Optional' if the page may have none.
get_sourceNEWSave the page's whole HTML to a variable (to find text / codes with 'Extract text by pattern').
More detail
Use to find hidden text in the page (token, id…) that cannot be clicked. Combine with 'Extract text by pattern' to get the exact part.
zoomNEWChange the page zoom (fit a small screen, or show more content).
More detail
Tip: small account windows (grid of many) → zoom 80% to see more. Element clicks still work; for COORDINATE / IMAGE clicks set the zoom first, then capture templates.
upload_fileNEWLoad a file from the PC into an upload field (<input type=file>), e.g. a video, pdf, specific image.
More detail
Target the file input (or the 'Add file' button — the tool finds the nearby file input). For a random image from a folder use 'Upload image from folder'.
highlightNEWRed border around matching elements for a few seconds so you can check the pick.
More detail
Use while writing a script: pick the element → test this command → see which element gets a red border in the browser. Check only, changes nothing.
Condition · 11 commands
if_countNEWCount matching elements then compare (e.g. at least 3 → continue).
More detail
Example: count 'Next' buttons — if more than 0, click. Counts instantly (no wait).
if_fileNEWCheck whether a file exists, has content, or contains some text.
More detail
Example: If file 'da_lam.txt' contains the account name {name}, skip this account (with 'End script for account').
Relative paths start from the tool's folder.
if_timeNEWIf the current time is within the window (e.g. 08:00 → 22:00) and on the right weekday, run the commands below.
More detail
Make accounts work only at certain hours, or rest at night. Windows across midnight are fine (e.g. 22:00 → 06:00).
Blank 'Weekdays' = every day. Numbers: 1 = Monday … 6 = Saturday, 7 = Sunday; e.g. 1,2,3,4,5 = weekdays.
if_existsNEWIf the element is found, run the commands below up to 'Else' or 'End if'.
More detail
If the element IS found (within the wait), run the commands below up to 'Else' or 'End if'. Great for optional popups: If 'Close' button exists → click Close → End if. Always close the block with 'End if'.
if_missingNEWIf the element is NOT found (after the wait), run the commands below.
More detail
Run the commands below when the element is NOT found (after the short wait you set). E.g. no 'Home' button → do the login steps. Close the block with 'End if'.
if_urlNEWIf the current URL contains this text.
More detail
Run the commands below when the current URL contains your text. E.g. contains 'login' → we are on the login page.
if_titleNEWIf the tab title contains this text.
More detail
Run the commands below when the tab title contains your text.
if_textNEWIf the element's text contains this.
More detail
If the element's text contains your text, run the block below. E.g. a notice containing 'success' → write OK to file; 'Else' → write ERROR.
if_varNEWCompare 2 values, e.g. {$tien} greater than 10.
More detail
Compare 2 values (usually variables) to branch. E.g. {$sodu} greater than 10 → continue, else skip. Numbers and text both work.
elseNEWCommands that run when the condition above is FALSE.
More detail
The 'ELSE' part: commands below this run only when the 'If…' condition above is FALSE. Optional.
end_ifNEWCloses an 'If' block.
More detail
Marks the end of an 'If…' block. Every 'If…' needs exactly one 'End if'; if missing the tool reports an error before running.
Data & storage · 14 commands
json_getNEWFrom a JSON text (usually the result of 'Call API'), take one item by path such as data.user.name or items[0].id.
More detail
JSON is the data format pages / APIs usually return, like {"data": {"name": "An", "list": [5, 6]}}.
Path: level names separated by dots; the Nth list item is [N] (from 0).
Example: data.name → An data.list[1] → 6
The 'JSON' box usually holds {$phanhoi} (variable from Call API). If the item is a list / object the result is JSON text.
kv_waitNEWWait until a store key meets a condition (present / equals / greater…). Lets accounts wait for each other.
More detail
Accounts run in parallel but sometimes must wait for each other: account A writes store key 'xong_A' = 1 when done, account B uses this command to wait until that key = 1.
Times out with an error (or continues if you choose so).
excel_setNEWWrite a value into a column, on the current account's row of the profile Excel file (e.g. 'Done', wallet, balance…).
More detail
Excel mode only. The tool remembers which Excel row each account came from and writes to that row.
• Column name: the text in the header row (row 1). A missing column is added on the right.
• CLOSE the Excel file before running (cannot write while open in Excel).
• Only .xlsx / .csv. Needs openpyxl for .xlsx (pip install openpyxl).
Example: Column = Status, Value = Done → each finished account writes 'Done' to its row. Back up the file before the first run.
regex_extractNEWSearch a piece of text and pull out exactly what you need: 6-digit code, email, link, number… saved to a variable.
More detail
Use when you can read a WHOLE passage but only need part of it. E.g. an email says 'Your verification code is 482913, expires in 10 minutes' → pick '6-digit code' → 482913.
Where the text comes from (top priority first):
① Pick an element in 'Pick element' → its text
② Or fill 'Text to search in' ({$variable} allowed, e.g. {$noidung} from Get value)
③ Both blank → scan all visible page text.
'Custom' is for regex users; with a ( ) group, the group is taken. E.g. Mã:\s*(\d+) takes the digits after 'Mã:'.
Use the result later with {$variable}.
text_opNEWTransform text: trim spaces, change case, replace, split by separator, take a slice, strip Vietnamese accents.
More detail
Common examples:
• From '[email protected]|password' → Split by | take part 1 → '[email protected]'.
• 'Nguyễn Văn A' → 'Strip Vietnamese accents' → 'Nguyen Van A'.
• Dot to comma → 'Replace': find '.' replace with ','.
• First 4 characters → 'Take a slice' from 1 to 4.
Input text may use {$variable}, {@Column}. The result is saved to a variable for later commands.
kv_saveNEWSave a value under a key in the store file; readable in later runs. Can be used for counting.
More detail
Normal variables ({$name}) vanish when the account finishes. The 'store' is kept forever in a file, shared by all accounts. Examples:
• Remember each account's last task date: key = ngay_{name}, value = {$today}.
• Count accounts done: key = tong, mode = 'Add number', value = 1.
Include {name} or {stt} in the key for one slot per account. Read back with 'Read from store'.
kv_loadNEWRead back a value saved with 'Save to store' (by key) into a variable.
More detail
If the key was never written, 'Default value' is used. Combine with If: e.g. read ngay_{name}, if it equals today skip this account.
reportNEWAppend one row to a csv report (opens in Excel); a new file gets a header row and a time column.
More detail
Unlike 'Append row to csv': it creates a header row for a new file and adds a Time column first. E.g. at the end of the script per account:
Columns: {name}|{$user}|{$code}|OK
Header: Acc|User|Code|Status
Columns separated by |. Parallel accounts write correctly without overwriting each other.
ptr_resetNEWMove a text file's 'position cursor' back to the line you choose (usually line 1).
More detail
When 'Send text' uses 'Cursor remembers position', the tool remembers which line of the file was used, even after closing and rerunning the next day. To start over run this (set line 1) — usually run ONCE separately, not inside a per-account script (it would reset constantly).
set_varNEWSet a variable's value (can combine {1}, {$other}).
More detail
Create/change a VARIABLE (a named box for data). E.g. name = trangthai, value = OK, later commands use {$trangthai}. Values can be combined: {name}_{$so}.
calcNEWCompute + - * / % and brackets, e.g. {$a}+{$b}*2. Result into a variable.
More detail
Simple math + - * / % and brackets. E.g. {$a}+{$b}*2. Result stored in a variable. For counting (dem = {$dem}+1) or percentages.
jsNEWRun arbitrary JS (for coders).
More detail
Run a snippet of JavaScript on the page — for programmers only. Beginners should use the built-in commands; to read a result use 'Get value from JS'.
save_txtNEWAppend a line to a txt file. Use {name} {stt} {$variable}.
More detail
Append one line to a .txt file. For saving results such as created accounts. File name may use {stt}, {name}. {$variable} allowed.
save_csvNEWAppend one csv row, columns separated by |. Opens in Excel.
More detail
Append one row to a .csv file, columns separated by | (e.g. {name}|{$ma}|Done). Opens in Excel. For an automatic header + time use 'Write csv report'.
Get value (screen) · 5 commands
d_get_screenNEWSave screen, browser window and viewport sizes to variables.
More detail
Creates variables (base name 'scr'): {$scr_sw} {$scr_sh} = screen width/height; {$scr_ww} {$scr_wh} = browser window width/height; {$scr_wx} {$scr_wy} = window top-left; {$scr_vw} {$scr_vh} = viewport width/height (where x, y coordinates apply).
For size-based coordinates, e.g. click the page center: x = {$scr_vw} / 2 (via Calculate).
d_get_colorNEWRead the pixel color at (x, y) on the page into a variable as #RRGGBB.
More detail
Check a button's state by COLOR: e.g. green = ready, gray = not yet. Read the color at the button's center, then compare with 'If value / variable'. The 🎯 button gets x, y.
d_find_posNEWFind the template on the page, save its center into {$name_x} and {$name_y} (no click).
More detail
Use to learn where an image is before acting: e.g. find the 'search box', then 'Click coordinates' with x = {$o_x} and y = {$o_y}, or 'Move mouse by offset'. Variable name 'o' → two variables {$o_x} and {$o_y}.
d_ocrNEWCapture the page / a region / an element and 'read' the text in the image. For text inside images that cannot be copied.
More detail
OCR = the machine looks at an image and copies the text. Useful for digit captchas, codes in images, balances drawn on canvas.
Where to read: the visible part, a region (x1,y1,x2,y2) or a picked element. Smaller region = more accurate.
One-time install: open CMD and run pip install rapidocr-onnxruntime (plain letters / digits read very well; accented Vietnamese less so). 'Keep only' drops extra characters, e.g. digits only.
The result is saved to a variable, reuse with {$name}.
d_clipNEWRead the clipboard into a variable, or put text on the clipboard (like Ctrl+C).
More detail
Typical use for text that cannot be picked: Click the field → Send key ctrl+a → Send key ctrl+c → this command 'Read' → text is in the variable.
Important: the clipboard is shared by the WHOLE PC, parallel accounts will overwrite each other. Use only with 1 account at a time (Parallel = 1). One-time install: pip install pyperclip.
Image · 7 commands
d_click_imgsNEWFind every match of the template on the page and click them top to bottom (e.g. every identical Follow button).
More detail
Unlike 'Click image' (best match only), this clicks ALL matches, top to bottom, left to right.
E.g. 8 identical 'Follow' buttons: capture one with ✂ → 'Max' = 5 → the tool clicks the first 5 with random pauses.
Searches only the visible part (scroll first). Background mode only. Several templates: separate with |. No match = error.
d_shotNEWSave an image to file: visible part, full page, a region, or one picked element.
More detail
Capture type:
• Visible part: like a screenshot of the page.
• Full page: top to bottom (including the scrolled part).
• A region: enter x1,y1,x2,y2 (px from top-left). • An element: pick it in 'Pick element' (e.g. save just the captcha image).
File name may use {stt} {name} {@Column}, e.g. shots/{stt}_{name}.png. To use the path later, set a variable name and call {$name}.
d_click_imgNEWFind the template image on the account's page and click it.
More detail
Find a TEMPLATE IMAGE on the page and click it. How: press ✂ 'Capture' → box the button/image → the tool saves it as a template. 'Similarity' 0.85 is fine; if the site changes colors lower it to 0.75 or pick 'Grayscale'. A small 'Search region' is faster and safer; 'Offset X/Y' clicks away from the image center (e.g. top corner).
d_move_imgNEWFind the template image and move the mouse there, no click.
More detail
Find a template image on the page and move the mouse there, no click. Make the template with ✂ (capture a small region). Several images separated by |.
d_wait_imgNEWWait until the template image appears (or disappears).
More detail
Wait until a template image APPEARS (or DISAPPEARS) on the page. Useful when the site has no stable xpath, e.g. wait for the loading spinner to vanish before continuing.
if_imgNEWIf the template image is found, run the commands below (up to 'Else' / 'End if').
More detail
If the template image is found, run the block below (up to 'Else' / 'End if'). E.g. popup X image found → Click image to close it.
if_noimgNEWIf the template image is NOT found, run the commands below.
More detail
Opposite of 'If image exists': run the commands below when the template is NOT found. Use with 'Else' and 'End if'. E.g. no 'logged in' image → go log in.
Browser control · 4 commands
d_win_setNEWChange window size / position, maximize, minimize, full screen.
More detail
Pick 'Action':
• 'Set size & position': enter x, y (top-left), width, height (px).
• Maximize / Minimize / Full screen / Restore.
Note: with several accounts running and 'Arrange windows' on, this command overrides that layout for the current account.
d_webNEWBrowser-level actions: go to URL, reload, back, forward, open / close tabs.
More detail
All browser actions in one command, chosen in 'Action'.
• Default mode (no real mouse): the tool drives the browser directly, in the background, many accounts at once.
• 'Take over real mouse' mode: the tool presses keys like a real user (Ctrl+L type URL Enter, F5, Alt+←, Ctrl+T, Ctrl+W), one account at a time, window must be visible.
Example: Action = Go to URL, URL = https://x.com. Include https://.
in_connectNEWAttach Playwright to the open browser for INDIRECT commands (XPath, CSS, JS...). Skipped if already attached.
More detail
Use after DIRECT opened the site and passed the easily-blocked steps. From here INDIRECT commands work.
The tool also connects automatically at the first INDIRECT command, so this is only for connecting early / explicitly.
To go back to 'real human' mode use in DisConnect.
in_disconnectNEWDetach Playwright from the browser; the page stays open. Then use only DIRECT commands until reconnected.
More detail
Use before easily-detected steps (login, captcha, buttons needing a 'clean' mouse). Reconnect with in Connect or any INDIRECT command.
Flow · 17 commands
log_msgNEWPrint a custom message to the Log (e.g. 'Account {name} reached home'). Can also append to a file.
More detail
Shows where the script is, or records results. May use {name}, {stt}, {$variable}, {@Column}.
Colors: green = success, yellow = warning, red = error (visual only, a red line does NOT stop the script).
Fill the file path in the last box to also append a timestamped line to that file.
auto_restNEWEach time this runs there is an X% chance of a random pause — makes the script less machine-like.
More detail
Sprinkle this through the script (e.g. between actions). With 30% chance, pause 3–15 seconds: about 3 of every 10 passes pause a little.
100% = always pause (like Wait with a random range).
repeatNEWRepeat the commands below N times, up to 'End loop'.
More detail
Example: scroll 5 times → 'Repeat N times' (N = 5), the scroll commands below, then 'End loop'. Everything BETWEEN runs N times.
repeat_forNEWRepeat the commands below until time runs out (e.g. browse 5 minutes), up to 'End loop'.
More detail
Unlike 'Repeat N times', you set a DURATION. E.g. 'simulate reading news for 3 minutes' → Repeat for 180 seconds, inside: wheel + wait 2-5 seconds, then 'End loop'.
Fill 'To' so each account differs (e.g. 120 → 300 seconds = random 2 to 5 minutes). When time is up the current pass finishes then exits. Exit early with 'Break loop'.
end_repeatNEWCloses a 'Repeat' block (both N times and N seconds).
More detail
Marks the end of a loop block (Repeat N times / Repeat for N seconds). Commands between the loop and this are repeated.
breakNEWExit the innermost loop.
More detail
Exit the innermost running loop right away and jump past 'End loop'. Use with 'If': once the condition is met, break.
stopNEWStop the current account's script (counts as success).
More detail
Stop the current account's script right here and count it as SUCCESS (not an error). Use with 'If': e.g. if this account already ran today, stop instead of repeating.
abort_accNEWThe account hit something it cannot pass (e.g. 'Confirm you're a human'): skip it, write one report line, close the profile, run the next account.
More detail
Use inside an 'If element exists' block to handle accounts blocked midway.
Example: after Send Text (paste mail) add:
① If element exists text=Confirm you
② Skip this account (reason: asked Confirm you're a human)
③ End if.
When reached the tool: writes one report line (if a file is set), CLOSES the account's profile (if Yes), marks it 'Skipped' in the table, then runs the next account. Other running accounts are unaffected, even if 'On error' is set to stop. The report may use {STT} {name} and saved variables like {$mail}. Only applies in the Command tab (not the Run Code tab).
restart_privateNEWClose the account's Firefox, reopen the same profile in private mode (-private) and rerun the script from the start. {$priv} = 1 during the private run.
labelNEWMarks a position in the script for 'Jump to label' to return to.
More detail
Like planting a flag. E.g. set label 'dang_nhap' at the start of the login part; if an error shows at the end, 'Jump to label' dang_nhap to retry. Label names: no spaces, no accents.
goto_labelNEWJump to a labeled position (up to retry, or down to skip a section).
More detail
Usually placed INSIDE an 'If…' block (jump only when true). Example:
① Set label 'thu_lai'
② … registration steps …
③ If text 'Error' → Jump to label 'thu_lai' → End if.
'Jump at most' prevents infinite loops: after that many jumps the tool skips the jump and continues.
rand_startNEWEach account draws one of the branches (separated by 'Else') to run, up to 'End branch'.
More detail
Makes accounts behave DIFFERENTLY, more natural. Structure:
Random branch — start
(branch 1 commands)
Else
(branch 2 commands)
End branch
Each account runs EXACTLY ONE branch. To favor branch 1: set weights, e.g. 70,30 (branch 1 70%, branch 2 30%). Blank = equal.
rand_orNEWSeparator between branches in 'Random branch'.
More detail
Separator between two branches in 'Random branch'. Commands between 'start' and this are branch 1, after it branch 2…
rand_endNEWCloses a 'Random branch' block.
More detail
Marks the end of a 'Random branch' block. Each run picks ONE branch then jumps past this.
call_scriptNEWRun a whole saved script (.json file) then come back and continue here.
More detail
Split long scripts: write 'dang_nhap.json', 'lam_nhiem_vu.json'… separately and chain them with this.
① Save the child script as .json (Save button in the Script tab).
② Pick that file here.
The child SHARES variables ({$name}) with the main script. Up to 5 levels deep.
pause_hereNEWStop here, handle it by hand (e.g. solve a captcha) then press ▶ Continue.
More detail
Use when a step cannot be automated (hard captcha, phone verification). The tool pauses ALL running accounts and writes a note to the log; do it by hand then press ▶ Continue on the top bar. Fill 'Note' to remember why it stopped.
popup_guardNEWFrom here on, before EVERY step the tool auto-clicks annoying popup buttons (Accept, Đồng ý, Close…).
More detail
Many sites show cookie banners / ads / notices that cover the button you need. Enable this once at the start, listing the button texts to auto-click, one per line, e.g.:
Accept all
Đồng ý
Close
Before every later command the tool quickly scans: any button/link with matching text is clicked. To stop, add this again with Off.
Note: specific text is safer (avoid very short text like 'OK', it may hit the wrong button).
File · Network · Wallet · 10 commands
file_opNEWWrite, append, read, delete, copy, move files; create folders; check existence; count lines; list files.
More detail
Pick one 'Action' each time:
• Overwrite / Append: save 'Content' to the file (Append = add at the end).
• Read: the whole file's text into a variable.
• Delete: deletes a FILE (folders cannot be deleted, for safety).
• Copy / Move: from 'Path' to 'Destination path'.
• Create folder.
• Exists?: variable = 1 if yes, 0 if not.
• Count lines: non-empty lines.
• List: file names in the folder (one per line).
Relative paths start from the tool's folder. {name}, {$variable} allowed.
run_cmdNEWRun a Windows command (or open a program, run a .bat / .py) and capture its output.
More detail
FOR THOSE WHO KNOW A BIT ABOUT COMMANDS. Examples:
• python tinh.py {name} → runs your Python file, output saved to a variable.
• notepad → opens Notepad.
Runs in the tool's folder. Killed on timeout. WARNING: the command runs exactly as written — only run commands you understand, never paste unknown ones.
http_requestNEWSend a request to a URL (GET / POST…) and save the response to a variable — fetch data, call an IP-rotation API, get OTP from a service…
More detail
Unlike 'Go to URL' (opens the page in the browser), this calls silently, no page shown. The response (usually JSON) is saved to a variable; cut it with 'Get value from JSON'.
• GET: fetch only. POST: send data (in 'Request body').
• Headers: one Name: value per line (e.g. Authorization: Bearer abc).
• Status code (200 = OK) in {$name_ma}.
Only call services you trust. The request goes straight from your PC (not through the account's proxy).
wallet_genNEWCreate MANY Ethereum/EVM wallets at once (address + private key + phrase), or paste an existing private key / phrase to derive the address.
More detail
EVM wallets work on Ethereum, BNB Chain, Base, Arbitrum… Each wallet is exported with 3 columns: address · private key · phrase (12/24 words).
• Create new: set 'Number of wallets' (e.g. 50) → 50 different wallets, each with address + private key + phrase.
• From phrase: paste phrases, ONE WALLET PER LINE → address + private key (+ phrase kept).
• From private key: paste 0x… keys, ONE WALLET PER LINE → address. NOTE: a private key can NOT be turned back into a phrase (the phrase makes the key, not the reverse — like grinding meat, you cannot un-grind it). The phrase column says 'none'. Key-imported wallets work normally.
Saved variables (base name 'vi'): {$vi_addr} {$vi_key} {$vi_mn} = FIRST wallet; {$vi_ds_addr} {$vi_ds_key} {$vi_ds_mn} = full list (one wallet per line); {$vi_so} = wallet count.
SECURITY: private key and phrase = full control of the wallet's funds. The Log only writes addresses, NEVER keys/phrases. Keep the exported file safe, never send it to anyone.
Needs the eth-account library (pip install eth-account).
wallet_setupNEWDownloads the wallet extension (OKX / Gate / Rabby / Bitget / Phantom) or uses a pasted extension folder, installs it into accounts and imports the wallet by seed phrase. Works for Firefox and Chrome (GPM) accounts.
More detail
One command does both; 'Action' lets you install only or import only.
• FIREFOX: downloads the .xpi from Firefox Add-ons, installs via Marionette (usable at once) or drops it into the profile (GPM BiDi accounts: reopen once).
• CHROME (GPM): downloads the .crx from the Chrome Web Store, unpacks into extensions/ next to the tool, adds --load-extension when opening accounts. Close already-open accounts so the tool can reopen them.
• PASTE FOLDER: 'Extension folder' takes an unpacked folder (e.g. E:\profileGPM\<acc>\Default\Extensions\<id>\<version>, or just the <id> folder). The tool copies a clean copy and loads it into EVERY Chrome account of the run, keeping the original ID. Pick 'Other extension' for non-wallet extensions.
• 'Use proxy?' = No: this run assigns no proxy, Chrome opens with --no-proxy-server, Firefox disables its proxy.
• Only accepts extensions with the right name + ID (against fake wallets). Check first: python gpm_wallet.py check okx gate rabby bitget
• Import: opens the wallet UI in a temp tab, reads the screen text to know the step (Import → phrase → password → confirm), never clicks 'Create'. Closes the temp tab when done.
• Phrase sources are the same as Send text (pasted list: line 1 → first ticked account…). Phrases must have 12/15/18/21/24 words. Phrase + password are NOT written to the Log.
• If the wallet gets stuck, the Log prints the visible text and buttons for tuning.
SECURITY: the seed phrase lives in the script / Excel file — use burner wallets only, never share a script with phrases filled in.
x_socialNEWGo to x.com: log in if needed (user / pass / 2FA), then browse the feed 3-4 minutes like a human and like a few random posts.
More detail
• 'Action' = Auto: logged in → just browse; otherwise log in then browse.
• Accounts: Excel columns (blank name → auto find User / Pass / 2FA / Mail columns, preferring ones named X or Twitter), or a pasted list / text file with user|pass|2fa|mail per line.
• 2FA: enter the secret key (long string), the tool generates the 6-digit code. If X asks for a captcha / email code the tool waits for you.
• Browse: varied scroll speed, reading pauses, occasional scroll-back, opens 1-2 posts then returns. Likes skip ads, at least 12 seconds apart.
• 'Action' = Log out: the tool clicks the real Log out on x.com (X ends the session), then clears x.com / twitter.com cookies + data in the profile. Cookies of other sites (Gmail, Discord...) are KEPT. No account needed.
• 'Wipe level' = Full wipe (default): clears cookies, site storage, IndexedDB, cache, service workers of x.com / twitter.com / twimg.com / t.co. Already logged-out accounts are wiped WITHOUT opening x.com (opening would write new cookies); ends on a blank page. Full wipe also removes X passwords saved by Chrome (so the login form stops auto-filling).
• 'Action' = Get auth_token: the tool reads the profile's X cookies, takes auth_token (+ ct0, twid) and writes a CSV (one line per profile: no, profile name, auth_token, ct0, twid, time). No account needed, runs in bulk and multi-threaded like other commands. Profiles not logged into X get a blank auth_token. Default file du_lieu/x_token.csv, opens in Excel.
• 'Action' = Log in with auth_token: the tool injects the auth_token cookie (plus ct0, twid if present) into the profile and opens x.com to check — logged in WITHOUT user / pass / 2FA. Token from: an Excel column ('auth_token column', blank = auto find a column named auth/token; ct0 / twid columns auto detected too), or a pasted list / text file with one auth_token per line (auth_token|ct0|twid to include them). OR quick: PASTE one auth_token in the 'paste directly' box (used for every ticked profile, overrides Excel/list) — handy for testing one account. Expired / wrong tokens fail that account. Enable 'clear old X cookies first' to avoid mixing sessions.
• Afterwards: {$x_likes} (posts liked), {$x_login} (san / moi / chua / token), {$x_logout} (ok / san = was not logged in), {$x_auth} / {$x_ct0} / {$x_twid} (tokens read) and {$x_settoken} (ok / loi / thieu for token login).
SECURITY: passwords and 2FA keys are never written to the Log.
geetest_v4NEWSend captcha_id to NoCaptchaAI, get a GeeTest v4 token into a variable. Runs in Python so no page CORS issues.
More detail
HOW TO USE:
① Create nocaptcha_key.txt next to the tool and paste the API key in it (not in the script).
② Leave captcha_id (32 hex chars) BLANK to auto detect; if that fails take it from the Network tab (request containing load?captcha_id=...).
③ The token is saved to {$variable}.
NOTE: this command does NOT retry (each call costs credit). Timeout / wrong key / no credit shows a clear error in the Log.
How to inject the token depends on the site; NoCaptchaAI docs are unclear — send the log line + site name for adjustments.
geetest_dxNEWRead the slide captcha image on the page, find the right gap, save the px to drag to a variable. Use with 'Drag element by offset'.
More detail
HOW TO USE (3 commands in a row):
① This command (once the captcha is shown) -> saves the px into {$geetest_dx}.
② 'Drag element by offset (captcha slider)': pick the SLIDER KNOB (🎯), dx = {$geetest_dx}, dy = 0.
③ 'Wait for element' with text 'Success' to confirm.
Slightly off: adjust 'Multiplier' (e.g. 1.03) or 'Extra px' (e.g. -3). The log prints the measurements.
hcaptchaNEWFinds the hCaptcha on the open page, sends it to YesCaptcha, injects the token and calls the site's callback. Token saved to {$hcap}.
More detail
WHERE TO PLACE: right after reaching the hCaptcha page (ideally with the form filled), BEFORE clicking Sign up / Submit.
CLIENT KEY: from yescaptcha.com. Safest is to put the key in du_lieu/yescaptcha_key.txt (one line) and leave 'Client key' BLANK, so sharing the script does not leak it. {@ExcelColumn} also works here.
SITEKEY: blank = auto detect (data-sitekey or in the hCaptcha iframe). If not found, F12 → search 'sitekey' and paste it.
PROXY: 'Account's current proxy' = YesCaptcha solves through the proxy the tool assigned (HCaptchaTask type, half price, token on the same IP as the browser). If the tool did not assign a proxy (GPM's stored proxy), pick 'Type manually' and enter {@Proxy} or that proxy; otherwise YesCaptcha's own IP is used.
THE PROXY MUST BE A PUBLIC IP: 127.0.0.1 / 192.168.x (local proxies) are unreachable for YesCaptcha → pick 'No proxy'.
THE TOKEN LIVES ~120 SECONDS and works once: press Submit as soon as possible after this. The command does not retry on error, to avoid paying twice.
Variables after running: {$hcap} = token, {$hcap_ua} = user-agent used by YesCaptcha, {$hcap_resp} = respKey (if any).
DISCORD-STYLE SITES (captcha appears in a dialog after pressing Login): add this command twice. First 'Prepare: set trap' BEFORE the Login click; second 'Solve now' after the captcha shows. The trap records the sitekey, rqdata and the site's callback, which the tool calls after solving. NO need to click the captcha checkbox (that only opens the image puzzle).
Note: this command reads the page, so in DIRECT mode the tool attaches Playwright for it automatically.
captchaNEWDetects the captcha type on the page (reCAPTCHA v2/v3, Turnstile, Cloudflare challenge, hCaptcha, GeeTest, text image, AWS WAF, DataDome, MTCaptcha), sends it to YesCaptcha or CapSolver and applies the result to the page. If one service fails it switches to the other. The result is stored in {$cap}.
More detail
WHERE TO PUT IT: after the step that opens the page with the captcha (ideally after filling the form), BEFORE the Submit / Sign up click. Cloudflare challenge page: right after opening the page.
KEYS: get them at yescaptcha.com and capsolver.com. Safest: write them to du_lieu/yescaptcha_key.txt and du_lieu/capsolver_key.txt (1 line) and leave the key boxes EMPTY. One key = one service; with both, a failure on one switches to the other.
WHO SOLVES WHAT: YesCaptcha = reCAPTCHA v2/v3, Turnstile, Cloudflare challenge, hCaptcha, text images. CapSolver = reCAPTCHA v2/v3, Turnstile, Cloudflare challenge, GeeTest, text images, AWS WAF, DataDome, MTCaptcha (CapSolver does NOT solve hCaptcha).
CAPTCHA TYPE: 'Auto-detect' is enough. If a page has several kinds, pick one by hand.
SITES THAT SHOW THE CAPTCHA AFTER A CLICK (dialog): add this step twice — first 'Prepare: set the trap' BEFORE the click, then 'Solve now' after the captcha appears.
reCAPTCHA v3: if the site uses its own action and the trap cannot catch it, press F12, find grecaptcha.execute(..., {action: 'xxx'}) and fill the Action box.
TEXT IMAGE CAPTCHA: fill 'Captcha image' (CSS/XPath of the image) and 'Text input box'. The tool reads the image, has it solved and types the text into the box.
CLOUDFLARE CHALLENGE / DATADOME: the account's public proxy is REQUIRED. The service gets the cookie (cf_clearance / datadome) through that proxy, the tool loads it into the browser and reloads the page. Cloudflare only accepts the cookie with the same IP and the same user-agent: CapSolver uses the browser's own user-agent (Chrome Windows/Mac), YesCaptcha always solves as Chrome Linux so some sites reject it — that is why the tool tries CapSolver first for this type.
PROXY: 'Account proxy' = solve through the account's own proxy (token from the same IP, less suspicious). If the proxy fails it retries from the service's IP (except types that require a proxy). 127.0.0.1 / 192.168.x proxies are unreachable for the service and are skipped.
A TOKEN LIVES ~2 MINUTES and works once: click Submit right after this step. The step is not retried on error, so you never pay twice.
VARIABLES: {$cap} = token or recognized text, {$cap_ua}, {$cap_loai}, {$cap_dv}. GeeTest v4: {$cap_lot_number} {$cap_pass_token} {$cap_gen_time} {$cap_captcha_output}.
Real mouse & keyboard · 8 commands
d_clickNEWMove to the element and click with a human-like path. Does NOT take over the real mouse by default (background).
More detail
'Human-like' click: curved mouse path, slightly off-center, random speed. Runs in the BACKGROUND by default without touching your mouse, so many accounts can run at once. Pick 'Take over real mouse' only when the site detects the virtual mouse (rare); then only 1 account runs at a time.
d_moveNEWMove the mouse to the element, no click.
More detail
Move the mouse (human-like path) to an element WITHOUT clicking. For hovering menus or natural behavior before a click.
d_click_xyNEWClick at point x, y on the page (px from the top-left of the viewport).
More detail
Click a point (x, y) on the page when no element can be picked (e.g. an image, canvas, game). The 🎯 'Pick point on page' button gets the coordinates: click the exact spot. 'Random offset' shifts each click a few px for a natural feel.
d_move_xyNEWMove the mouse to point x, y on the page.
More detail
Move the mouse (human-like) to point x, y, no click.
d_typeNEWType key by key like a human (Vietnamese supported). If an element is set, it is clicked first.
More detail
Type key by key like a human, Vietnamese supported. If an element is set, it is clicked first. With 'Typos' on (e.g. 4%) it sometimes hits a neighbor key and corrects it — more human. Passwords/keys are hidden in the log.
d_pressNEWReal key press: enter, tab, esc, ctrl+a, ctrl+v...
More detail
Press keys like a real keyboard: enter, tab, esc, or combos ctrl+a, ctrl+v, shift+tab. After typing in a search box (enter) or to move between fields (tab).
d_scrollNEWTurn the mouse wheel on the page. Positive = down, negative = up.
More detail
Turn the mouse wheel: positive = down, negative = up, 5 notches ≈ half a page. Fill 'To (random)' so each turn differs (e.g. 3 to 8 notches) — more like a reader.
paste_textNEWPaste the whole text at once (fast, like Ctrl+V), unlike typing key by key.
More detail
Use for long content (sentences, paragraphs) or odd characters/emoji that break key-by-key typing. Pick an element to click before pasting; blank = paste at the cursor.
Content may use {1} {$variable} {@ExcelColumn}.
Advanced mouse · 7 commands
d_mouseNEWStep-by-step mouse control: PRESS, RELEASE, or click in place. For 'hold to verify' buttons and manual drag-and-drop.
More detail
When? When a site requires HOLDING the mouse (e.g. 'Press and hold to prove you are human'). A 3-second hold = 4 commands:
① 'Move mouse to element' (the button)
② This command, Operation = Press and hold
③ 'Wait' 3 seconds
④ This command, Operation = Release.
Blank x, y = act at the current mouse position (after Move mouse). With x, y = move there first.
Always pair 'Release' with 'Press and hold', or the mouse stays stuck down.
d_dragNEWPress at point A, drag smoothly like a human to point B, release.
More detail
For dragging jigsaw pieces, puzzle parts, reordering lists. Coordinates from the page's TOP-LEFT (px).
To get coordinates: 'Click coordinates' commands have a 🎯 'Pick point on page' button — click point A for (x, y), same for B.
To drag an element a computed distance (slider), 'Drag element by offset' is easier.
d_drag_elNEWPick the element (slider knob) and drag it right/left/up/down by dx, dy px.
More detail
For 'drag the slider to fit' captchas. Steps:
① Pick the SLIDER KNOB as the element (🎯 button).
② dx = px to drag right (negative = left), dy = 0 for horizontal only.
③ Unsure how many px? Measure on a screenshot: track length minus knob width. Try 200 first then adjust.
The tool drags smoothly with slight human-like wobble. Tip: add 'Wait for element' ('Success' text) afterwards to confirm.
d_move_byNEWFrom the current mouse position, move by dx, dy px (no click).
More detail
Example: after 'Move mouse to element', nudge 40 px right to avoid the button → dx = 40, dy = 0. Negative = left / up.
d_click_boundNEWClick a random point inside rectangle (x1,y1)-(x2,y2).
More detail
Use to click 'somewhere' without repeating the same spot (e.g. an image area, empty space to deselect).
Enter the top-left (x1, y1) and bottom-right (x2, y2) of the region, in px from the page's top-left.
d_click_colorNEWFind the color you specify on the page and click it.
More detail
Use when the target has no stable xpath but a distinctive COLOR (e.g. green button, red captcha dot).
① Color as #RRGGBB, e.g. #FF0000 = red, #00FF00 = green, #0000FF = blue. To find a color code: screenshot, open in Paint/Photoshop with the eyedropper; or search 'color picker' on Google.
② Tolerance: web colors drift slightly → use 30. Many similar colors on the page → tighten (lower tolerance) or set a 'Search region'.
③ Search region: x1,y1,x2,y2 (e.g. 0,300,800,700). Blank = whole page. The tool clicks the LARGEST color blob found.
d_activeNEWBring the current account's window to the front (when covered).
More detail
Most commands do not need this since they run in the background. Only for 'Take over real mouse' when the window is covered, or before a real screenshot.
No matching commands.