Using ActionWright

Working in the browser

How the browser agent uses its own browser, when it pauses for you, and what it will not do.

What the browser agent does

The browser agent is the part of ActionWright that uses websites for you. You don't start it yourself. When a task needs real work on a website, ActionWright hands that part to the browser agent, and the result comes back into your task. See How it works.

Use it to research, compare, read, find your way through a site, and fill in forms. For example:

  • "Compare the prices of these three headphones on the maker's site."
  • "Open the opening hours page of the city library and tell me when it is open on Sunday."
  • "Read the latest article on this blog and summarize it."
  • "Find the return policy on this shop's site."
  • "Fill in the contact form on this page with my name and question, but don't send it."

For simple lookups, ActionWright may search the web and read pages without opening the browser at all.

Before you begin

  • Install Google Chrome. The browser agent uses the Chrome installed on your PC. Without it, browser tasks fail with a message asking you to install Chrome.
  • Set up an AI provider. See AI providers.
  • If the task needs an account, you can sign in ahead of time. See Its own browser.

There is no setting to turn on. The browser agent is always available.

Its own browser

The browser agent works in its own Chrome window with its own profile. It is separate from the Chrome you use every day: it does not see your usual tabs, history, bookmarks or sign-ins.

  • Sign-ins are remembered. When you sign in to a site in the agent browser, that sign-in stays in its profile for later tasks, also after you restart ActionWright. Anything the agent does later runs with that account.
  • Watch it work. Click Show Agent Browser in the task window to open the agent browser or bring it to the front. See Show the agent browser.
  • Where a task starts. If the agent browser is already open, the browser agent starts from the page that is open there, unless the task needs another page. Otherwise it opens a new tab.
  • Idle. The agent browser does not close on its own when it is idle.

Sign in ahead of time

  1. In the task window, click Show Agent Browser.
  2. Go to the site and sign in, as you would in Chrome.
  3. Run your task. The browser agent uses that sign-in.

Sign out or clear the profile

There is currently no ActionWright setting to sign out of all sites or to clear the agent browser's profile. To sign out of one site, open the agent browser with Show Agent Browser and sign out on that site.

To sign out of every site at once, clear the whole profile:

  1. Right-click the ActionWright icon in the Windows tray and choose Quit.
  2. In File Explorer, go to %USERPROFILE%\.action-wright.
  3. Delete the browser-profile folder.
  4. Start ActionWright again. The next browser task starts with a new, empty profile that is not signed in anywhere.

This also removes the agent browser's history, cookies and site data. It does not affect your everyday Chrome. For everything ActionWright keeps on your PC, see Privacy.

When it hands control to you

Some steps stay with you. The browser agent does not type passwords, one-time codes, card or bank details, or ID numbers. When a page needs one of these, or something only you should decide, it stops and hands the browser to you. This includes:

  • Signing in.
  • CAPTCHAs and other "are you a robot" checks.
  • Two-factor codes.
  • Entering payment details.
  • Chrome permission prompts, for example location, camera, microphone, notifications or passkeys.

Complete the step and press Done

  1. Under the result in the task window, a message says what you need to do. If the task window isn't in front, Windows shows an Action required notification.
  2. Click Show Agent Browser if the agent browser isn't in front, and complete the step there.
  3. Optionally type a short note in Reply (optional)..., for example "I chose the second address".
  4. Click Done, or press enter in the reply box.

The browser agent then looks at the page again and continues with the steps it has left. A handoff has no time limit. To end the task instead, click Stop.

What asks first

Reading and finding your way around a site run without asking. For anything that commits something, such as sending, submitting, deleting or paying, your request decides what this task may do. See How ActionWright decides what may run and the full table.

  • Every browser approval is for that one action only. Always Allow is never offered for browser actions. The next action asks again.
  • Before you allow, read what the dialog shows, such as the site, the button, the recipients and the text that will be sent.
  • A site can see what is typed into its fields even before a form is sent. Don't ask the browser agent to type private information on sites you don't trust.

For the dialog and what each choice does, see Approvals.

Protections

  • Payments stay with you. ActionWright is built to ask before anything that pays or places an order. Card and bank details are entered by you. See When it hands control to you.
  • Your request sets the limits. Actions are checked against your request. One your request did not allow asks you first, and one that clearly has nothing to do with your task does not run, even if a page asks for it. See How ActionWright decides what may run.
  • No repeats. An action that was already done or that you denied in this chat is not done again in another way, unless you ask for it again.
  • Uploads. The browser agent can upload only files you attached to the chat. It cannot pick other files from your PC. See Attach files.
  • Local files. The browser agent does not open files from your PC as web pages.
  • Downloads. Files the browser agent downloads are saved in your Windows Downloads folder, under the site's file name, without replacing an existing file. They are marked as coming from the internet, the way Chrome marks downloads. The browser agent never opens or runs them. Only Computer Use works on your desktop, so it is the one part of ActionWright that could open a downloaded file during a task, and installing software asks first unless your request asked for it. To keep any agent from opening files on your desktop, turn Computer Use off.
  • Program downloads ask first. A download that could run a program asks before it is saved. If you deny it, it is not saved to your Downloads folder.

If a cookie banner blocks the page, the browser agent may accept it so it can continue.

Limits

  • Step limit. Each run has a step limit. When it is reached, ActionWright asks whether to continue. See Limits per run.
  • Robot checks. Some sites detect automated browsing and show a robot check. Complete it yourself. See Complete the step and press Done.
  • Google Chrome only. The browser agent does not use Edge, Firefox or other browsers.
  • One profile. All tasks share the same agent browser profile and its sign-ins.
  • Stop during an action. If you press Stop while the agent is sending or submitting something, that action may or may not have happened. Check the site before you continue. See Stop a task.
  • Forms can look filled when they are not. On some sites a field, such as a date picker, can show a value the site did not accept. Check forms the browser agent fills in before they are sent. A form you asked it to fill but not send is the safest.
  • Late downloads. If a download starts as the very last step of a task, ActionWright may not mention it. Check your Downloads folder.

What is sent to your AI provider

The browser agent works through your AI provider, so the pages it reads, including their text and screenshots, are sent to that provider. Values of password, payment and ID fields are hidden from the AI. See Privacy.

If something goes wrong

ProblemLikely causeWhat to try
A message says Google Chrome is requiredChrome is not installed.Install Google Chrome, then run the task again.
"Could not open browser: …"The agent browser could not start.Click Show Agent Browser again. If it keeps failing, quit and restart ActionWright.
The task waits with a message and DoneThe agent handed you a sign-in, check or choice.Complete it in the agent browser, then click Done.
A robot check keeps appearingThe site detects automated browsing.Complete the check and click Done. If it keeps coming back, do that part yourself.
The agent isn't signed inYou signed in in your everyday Chrome, not in the agent browser.Sign in with Show Agent Browser. See Sign in ahead of time.
An action was blockedIt did not match your request, or it was already done or denied in this chat.Say exactly what you want in a new message, for example "submit the form".
The agent can't upload a fileThe file isn't attached to the chat, or it is over an attachment limit.Attach the file with + and ask again.
A downloaded file is missingYou denied a program download, or the task ended right after the download.Check your Downloads folder, or ask again.
"Step budget reached" appearsThe job needed more steps than one run allows.See The step limit.
You clicked Stop during a send or submitThat action may or may not have happened.Check the site before you continue.

For other problems, see Troubleshooting.