Translate

Translates text into another language on the phone itself, offline and without an AI account, working out what language it is written in.

An action in the AI group. Its type is action.translate.

Declared ports

Port	Label	Flow	Type
in	in	execution in	—
out	out	execution out	—
text	Text	data in	String
translation	Translation	data out	String

Configuration

Field	Key	Default	Notes
Text	text	—	—
Translate from	sourceMode	detect	detect, chosen
Which language	sourceLanguage	—	chosen from TRANSLATE_LANGUAGE; shown when sourceMode is chosen
Translate to	targetLanguage	—	chosen from TRANSLATE_LANGUAGE
If it fails	fallback	—	—

Notes

Translate turns text into another language and puts the result on the translation port. It runs on the phone, so it needs no AI account, no key and no signal.

• Translate from defaults to working the language out from the text itself, which is what you want when the text arrives from a message, a notification or a transcript.
• Translate to has no default. A blank one is an error, not “the phone’s language”.
• Both fields offer only the languages you have added under Setup, Translation models. The node never downloads anything itself.

Working of Translate

• The node reads Text, which is wired far more often than typed. The text worth translating is usually the one a trigger just handed you.
• With Translate from left at Detect automatically, the text is examined and a language is chosen. Text too short to judge — a name, an emoji, OK — is reported rather than guessed at.
• Both languages are checked against what is on the phone. One whose model has been removed ends the run there, with a message naming it.
• The translation lands on translation. Every failure lands on If it fails and the node still pulses out.
• Text that is already in the target language comes back unchanged. That is a success, not a failure, and it works even with nothing added at all.

Example: read a foreign message aloud in your own language

• Open Setup, Translation models and add your own language plus the one you expect to receive.
• Add a Message Received trigger and pick the messenger you want.
• Add Translate and wire the trigger’s message text into Text.
• Leave Translate from on Detect automatically and set Translate to to your own language.
• Wire translation into a Speak node.

Explanation:

• Detection is doing the work here. You do not know in advance what language a message will arrive in, so a fixed source language would be wrong most of the time.
• Speak reads whatever it is given, so the phone says the message in your language while the sender wrote it in theirs.

Adding Languages

Translation runs on a model per language, and you choose which ones the phone keeps.

• Setup, then Translation models, then Add a language. Each is about thirty megabytes and downloads there and then.
• English is built in and is never listed. Everything else is a download.
• Translation goes through English, so German to French needs German and French added.
• The language chooser on the node links straight to that screen, so you can add one without hunting for it — but it leaves the editor, so add what you need before you wire a graph up.
• Deleting a language stops every macro that translates into or out of it, until you add it back.

Points to Remember

• The node never downloads. A macro that fires at three in the morning will not start a thirty-megabyte transfer on your mobile data, and will not sit waiting for Wi-Fi that never comes.
• Because the fields only offer what is added, a node you were able to configure is a node that can run. The one way to break it is to delete a language it was using — and then the message names that language.
• Translate to blank is an error the node reports. It never falls back to the phone’s language, because a macro would then do something different on somebody else’s phone.
• Detection needs something to work with. One or two words often cannot be judged, and the node says so rather than picking wrongly.
• Detection itself needs nothing added — it will happily name a language you have not installed, and the node then tells you that is what happened.
• Nothing here reaches the network at all. The translation never leaves the phone, which is the reason to use this rather than Ask AI.
• Nothing halts. A refusal lands on If it fails and out pulses anyway, so compare the output with an If node when a failed translation should change what happens next.