TL;DR:
A secret you type yourself never has to pass through the assistant, so it cannot be copied, logged or misused there.
When a task reaches a login page, the right move is to stop, tell you, and wait while you sign in.
DuckyHelper is built not to type your passwords or card numbers.
Say you ask an assistant to pull last month's invoices from your electricity provider's website. It opens the site, finds the right page, and then the page asks for an email and a password. The assistant can see the two empty boxes. The question is whether it should fill them.
The safest answer is no. Passwords and card numbers are the two things that turn a helpful tool into a risk if they go wrong, and the person who owns them is sitting right there. Letting that person type them costs a few seconds.
Why secrets are different from other text
Most of what an assistant handles is low stakes. If it types the wrong word in a note, you delete the word. A secret is different because it is the thing that proves you are you. Anyone or anything that holds a copy can act as you, often without you noticing.
Each kind of secret has its own twist:
Passwords are reused far more than they should be, so one leak can open several accounts.
Card numbers let someone spend your money, and getting it back means disputes and replacement cards.
What happens when an assistant types a secret
To type a secret, an assistant first has to have it. That means it was stored somewhere, or you told it, or it read it off the screen. Each of those places is another copy that can be recorded, shown in a history, sent along with a request, or captured in a screenshot.
The fewer places a secret exists, the fewer places it can leak from. When you type it yourself, straight into the page, the only copies are the one in your head and the one the website receives. That is the smallest footprint a password can have.
Walk-through: a login page in the middle of a task
Here is the invoice task done the careful way, step by step.
You ask: "Download my invoices from the last three months from my electricity account."
The assistant opens the site: It reaches the provider's sign-in page. It can tell this is a login form because of the labels and the password field.
It stops and says so: Something like: "This page needs you to sign in. I will not type passwords. Tell me when you are done." It does not guess, retry or look for a workaround.
You sign in yourself: You type the password in the page, approve any code sent to your phone, and tell the assistant you are in.
It carries on: Now signed in, it goes to the billing page, finds the three invoices and downloads them. You review the files.
Notice that the assistant never needed your password to do its job. It only needed you to be signed in. Those are very different requirements, and a good design keeps them separate.
The case for "but it would be faster"
It would, by roughly ten seconds. Storing a password to save those seconds means a stored password for every site the assistant ever works on. That is a bundle of keys in one place, and it becomes the most attractive thing to steal.
There is also a mistakes problem. An assistant working fast on a page can paste a card number into the wrong box or a code into a search field. A human glance catches that. A fast, confident assistant might not.
A faster way that stays safe: If you want sign-ins to be quicker, use your browser's or operating system's own password manager. It fills the page for you and the assistant never sees the secret.
What this does not solve
Keeping secrets out of the assistant's hands does not make every risk disappear. If you sign in on a page that looks right but is a fake, typing the password yourself will not save you. Check the address bar before you type, with or without an assistant.
It also does not mean an assistant can never use a signed-in site. Once you have signed in, it may be able to work inside that session. That is a reason to stay present, watch what it does, and sign out when you finish a sensitive job.
Privacy and control
A well-behaved assistant gives you three things here: a clear stop when a secret is needed, a clear signal when it has resumed, and an easy way to cancel. If any of those is missing, treat the tool with more suspicion.
DuckyHelper follows this rule. It's built not to type your passwords or card numbers, and it's built to ask first before it sends a message or pays for anything. Its features that do work in other apps are in early access, and this rule applies to them too. You can read more on the safety page.
Try it yourself
Next time any tool offers to log in for you, ask what it stores and where. Then try the split on your own: let the assistant do the clicking, take over at the sign-in box, and hand back the keyboard afterwards. It feels slower for the first minute and then stops registering.
Frequently asked questions
Is it safe to paste a password into a chat so the assistant can use it?
No. Treat anything you paste into a chat as something that may be stored. Type secrets only into the site that needs them.
Can an assistant still help me on a page with a login?
Yes. It can guide you to the sign-in page, explain what the site is asking, and continue once you are signed in.
Get started
