The direct answer
AI Translator Keyboard does not send ordinary keystrokes to Launchborn’s servers and does not keep a history of messages or translations. An AI action starts only after the user taps the sparkle key. In on-device mode, the requested translation stays on the phone. In your-own-key mode, the selected text goes to the provider configured by the user under that provider’s terms. In Pro mode, the selected text passes through an encrypted Cloudflare proxy only to return the requested result and is not stored by AI Translator Keyboard. Password handling is separate from ordinary text: Apple documents that iOS replaces a third-party keyboard with the system keyboard in secure fields. On Android, AI Translator Keyboard detects password input types, disables suggestions and autocorrect, and does not submit password text for translation. No keyboard vendor can responsibly promise that every third-party app labels every field correctly, so users should still confirm that sensitive fields display the expected secure-input behavior.
What a third-party keyboard can access
On iPhone, a custom keyboard runs as an app extension in a separate process. Apple provides a text-document proxy so the keyboard can insert, delete and inspect limited surrounding context needed for features such as autocorrection. Apple also places extensions in a sandbox. Network access for an iOS keyboard requires the user-granted “Allow Full Access” capability.
Android keyboards operate as input method editors. The current text field tells the keyboard what kind of input it expects, including password variations. A responsible keyboard uses that signal to disable learning, suggestions and cloud actions for sensitive input.
This is why a blanket claim such as “a keyboard can never see text” would be misleading. A functional keyboard necessarily handles the characters required to type. The meaningful privacy controls are purpose limitation, explicit actions, secure-field behavior, minimal network transfer and retention policy.
What happens in each translation mode
| Mode | Where requested text goes | Stored by AI Translator Keyboard? | Account required? |
|---|---|---|---|
| On-device | Translation runs locally on the phone after the model is available | No | No |
| Your own key | Directly to the provider selected and configured by the user | No; it does not pass through Launchborn’s proxy | No AI Translator Keyboard account |
| Pro | Through an encrypted Cloudflare proxy to the active model provider | No; processed in transit and discarded | No; store entitlement verifies access |
Cloud processing cannot occur without transmitting the requested content to a model provider. “Private” therefore should not be used as a synonym for “offline” unless the selected engine actually runs on-device. AI Translator Keyboard exposes the engine choice so the user can make that tradeoff deliberately.
Why password fields are different on iPhone
Apple’s developer documentation is explicit: secure text fields always show Apple’s system keyboard. If a third-party keyboard is active, iOS temporarily removes it as soon as the user begins entering text in a secure field, then restores it in a normal field. Apple also allows individual apps to disable third-party keyboards entirely.
Apple’s platform security guide explains the second layer: custom keyboards run in a restrictive sandbox by default, and Open Access requires user consent. These operating-system controls are stronger than a promise made only in marketing copy because they are enforced outside the keyboard app.
Users can verify the behavior themselves. Open a normal text field, select AI Translator Keyboard, then tap a correctly configured password field. The visible keyboard should switch to Apple’s system keyboard. If an app uses a plain text field for a password, that app has failed to request secure input; do not enter a real password there.
How password handling works on Android
Android sends the keyboard metadata describing the active editor. AI Translator Keyboard checks password input variations and enters a restricted password mode. Suggestions and autocorrect are disabled, typed password content is not added to the personal dictionary, and AI translation actions do not submit the password.
Android has many device vendors and apps, so implementation quality outside the keyboard matters. A login field should use a password input type and visually mask the value. If a suspicious app presents an unmasked ordinary field, the safest response is not to enter the credential.
For everyday messages, the same explicit-action rule still applies: normal typing remains in the app being used, and an AI engine receives text only after the user requests an AI operation.
A practical checklist before installing any AI keyboard
- Read the privacy policy. It should distinguish ordinary typing, on-device processing and cloud processing.
- Look for retention language. “Encrypted” describes transport; it does not say whether content is stored.
- Check account requirements. An account creates another identity and data surface.
- Verify password behavior. Test a secure field before entering a real credential.
- Prefer explicit AI actions. A visible tap is easier to understand than continuous background processing.
- Choose a local engine when sensitivity matters. On-device processing minimizes the number of systems that receive the text.
For complete legal and provider details, read the current AI Translator Keyboard Privacy Policy. For a product-level comparison, see AI Translator Keyboard vs Gboard vs SwiftKey.
Frequently asked questions
Does AI Translator Keyboard record every key?
No. Ordinary keystrokes are not sent to Launchborn’s servers. The keyboard processes the text needed for an AI action only after the user requests that action.
Can AI Translator Keyboard see my iPhone password?
In a correctly configured secure field, iOS removes the third-party keyboard and shows Apple’s system keyboard, so AI Translator Keyboard is not present for that input.
Does Full Access mean every message is uploaded?
No. Full Access permits network functionality; it does not itself upload text. What is transmitted depends on the keyboard’s implementation and the engine selected by the user.
Is cloud translation private?
Cloud translation necessarily sends the requested text to infrastructure that can produce the result. Privacy depends on transport security, purpose limitation, provider terms and whether the service retains the content.