Voice is the feature people ask the most careful questions about, so here's the whole story in one place, moved off the pricing page where it was asking shoppers to care about architecture before they cared about the product.
Voice that learns instead of asking every time
Voice commands are not one big call to a language model. Your phone reads the sentence first, checking it against what you have already written, so a shop you have typed before is recognized as that shop rather than as whatever it sounded like. Most everyday commands are finished there and never leave the device.
What your phone cannot place goes to us, and even then it usually costs nothing to answer.
What leaves your phone is stored as a shape
Sentences shaped like ones other people have said are matched against a pool of patterns that holds the shape and not the contents, with the name and the amount stripped out. Sentences you personally have said before are remembered against your account as a shape with the name and the amount taken out, filed under a one-way code rather than the words, so the second time is instant and there is nothing in that store anybody could read back.
And one short record that does hold the sentence
The pattern pool is not the only thing kept, and this page would be dishonest if it read that way. When a command runs, the sentence itself is written to a log against your account, and it deletes itself after 30 days. I keep it because it is the only way to find the commands that were heard wrong and teach the recognizer the words you actually use. It never appears in the app, it is not used for advertising, and deleting your account erases it immediately. If you want yours cleared sooner, write to info@buhata.com and I will do it. The privacy policy describes this same record in its voice section.
A thumb up or down on a command is kept with the rest of your Buhata data, sentence and all, along with the reasons you pick and whatever you type about it, so it's in your account's copy on the server while you're signed in. If the AI model or your own saved phrases read the command, the thumb reaches me when you press it, with the sentence that was answered. An up can save that reading as one of your phrases, and it counts toward that pattern joining the pool. A down forgets the reading and keeps the command's pattern, with the name and the amount taken out, against your account until you've confirmed that kind of command twice or you delete your account.
Ticking "Allow Buhata to use this request to improve commands" adds a separate record on top of that: the sentence, the reasons, which part of Buhata read it and the pattern it was read as, for the few people who fix commands, and that copy deletes itself after 90 days. What you type about it isn't part of that record.
Only the awkward ones reach a language model
Only the genuinely awkward ones go to a language model. That is a small share of what people say, and it is the reason the microphone does not have to be priced like an AI product.