What Juniper does with a voice note.

Talking is often the quickest way to get a day down. You can do it on a walk, or while dinner cooks, without finding a keyboard. The trouble starts when you read it back.

What talking gives you

A raw transcript of speech is hard to read. It keeps every “um”, every sentence you started twice, and every place you changed your mind halfway through. It can look like this:

so um the run today was, the run was fine, like I did the whole loop and um I didn’t stop at the hill which is the first time, the first time I didn’t stop at the hill

That’s hard to read back in a month, let alone a year. And before you record anything, it’s fair to ask where the recording goes and what happens to it there.

From record to entry

In YearCraft, a voice entry goes through these steps, in this order:

1. You tap record. If AI sharing is off, YearCraft asks first, with “Turn on sharing” and “Type it instead” side by side.
2. You talk. When you stop, the recording is saved to your account.
3. The recording goes to an AI provider, which turns the speech into text.
4. Juniper tidies the transcript. More on what that means below.
5. The words land in your entry, where you can edit them like anything you typed.
6. If you save without a title, Juniper writes one, two to six words long.
7. The recording stays with the entry.

On the web you record in the browser. The iPhone app is on its way, and it keeps recording when the screen locks, with pause and resume from the lock screen.

The recording plays back inside the entry, in the voice note player. It has a waveform you can click or drag to move around, a button that goes back 15 seconds, and a speed pill that steps through 1×, 1.5×, and 2×. You can download the recording, or remove it and keep the written entry as it is.

What Juniper changes, and what she leaves alone

Juniper’s cleanup does three things. She removes filler words and accidental repetition. She fixes punctuation and paragraph breaks. She keeps your own words and your meaning, so the entry still sounds like you.

If you asked a question out loud, it stays a question. Review the text afterward and correct anything that isn’t right.

Tidied, the example above might read like this:

The run today was fine. I did the whole loop, and I didn’t stop at the hill. That’s the first time I didn’t stop at the hill.

What goes to an AI provider

A voice entry sends two things. The recording goes to an AI provider so it can be turned into text, and the transcript goes so it can be tidied. If you save the entry without a title, its text goes too, so Juniper can write one. Nothing else from your journal goes with it.

Your writing is never used to train AI models, ours or the provider’s. The full list of what each feature sends, and which providers receive it, is in the “What gets sent” sheet in Settings and in the Privacy Policy.

Sharing is your choice. YearCraft asks once when you start, and you can change your answer in Settings whenever you like. With sharing off, writing, goals, and achievements all still work, and voice entries and Ask Juniper pause until you turn it back on. Turning sharing off stops new requests. It can’t reach back and recall what was already sent.

Limits on each plan

Each stored recording counts as one voice entry, and the month follows your own timezone.

• Free: 3 voice entries a month, up to 10 minutes each
• Pro: no monthly limit, with recordings up to 20 minutes each

Pricing has the rest of what each plan includes.

Dictation or a voice entry

Your keyboard’s mic does something different. Dictation turns speech into text as you talk, the way it does in any other app. It’s free on every plan, it doesn’t count toward any limit, and it stores no recording, so there’s nothing to play back and nothing for Juniper to tidy.

Use dictation when you want the words and nothing else. Record a voice entry when you want to keep the recording itself, your voice on that day, next to an entry you can read.