Aurora for iPhone

Below is Aurora's iPhone interface, reproduced. It is a scripted replica running a sample conversation — not a recording of the live app and not a screenshot of one — and it walks itself through the session this page describes. Touch it and it stops performing and lets you drive.

A session, end to end

1. Press record. Aurora listens through the microphone of the phone you are holding. Nothing dials into your meeting and nothing else is asked of the room.
2. Watch the transcript arrive in two zones. Confirmed words sit in the transcript; the newest words stand in a dimmed tail behind them and firm up as the recognizer stops changing its mind. You are always looking at what Aurora is sure of and what it is not.
3. See the speakers separated into lanes. A diarizer running beside the recognizer decides which voice a stretch of speech belongs to, and each voice gets its own lane and color in the transcript.
4. Get the names the conversation gives. When someone says who they are, the model reads that out of the transcript and puts the name on the lane; otherwise the lanes stay Speaker 1, Speaker 2. Either way the name belongs to that session and is not carried into the next one.
5. Stop, and read the write-up. A language model on the device produces the title, the summary and the action items. You can tick the items off, and you can edit the title and the summary by hand.
6. Ask Aurora about the session. Put a question to it and the answer comes back with the moment in this transcript it rests on, or with the admission that the evidence is not there.
7. Take the text with you. Share or save the transcript as plain text, to whatever app or place you pick.

Everything in that list happens on the phone. What Aurora does is the same set of capabilities written out with their status labels, and the Privacy Notice names every path that does use the network.

What you need

• An iPhone running iOS 18 or later. That is the app's own minimum, taken from the project it ships from.
• Apple's TestFlight app. The link below opens it, and the App Store will install TestFlight first if it is not there yet.
• Room for the speech models, and a connection to fetch them once. They are downloaded on first launch rather than bundled with the build, and after that the session work needs no connection.

We have not published a tested list of iPhone models, and we will not print one we have not measured. Newer iPhones do the on-device work faster and older ones do it more slowly, which is the honest shape of it.

One caveat we cannot settle from the source: an extension embedded in the app declares a later minimum system than the app itself, and only the version App Store Connect shows for the build can say what installing it actually requires. If TestFlight refuses your phone on iOS 18, that is the thing to tell us.

How to get it

Aurora is on TestFlight and free while the beta lasts. The link is public: there is no waitlist, no account with us, and nothing to request.

Open in TestFlightEmail me the link

Or copy the link: https://testflight.apple.com/join/NpdW2ZEb

Open it on the iPhone you want to record with. The beta page carries the installation steps and what the beta covers; there is no second way in.

What this build cannot do

Written against the build on TestFlight today, with the label that says how far along each one is.

• Languages other than English. Experimental in beta: they work unevenly, and we have no per-language measurement to publish.
• Sessions much longer than an hour. Experimental in beta: they run, and they cost more time on older hardware.
• Recognizing a voice from an earlier meeting. In development: a name lives with the session it was found in, not as a voiceprint, so the next conversation starts from unnamed lanes.
• Searching across your conversations. In development: search works inside a session, not across the library.
• Syncing conversations to another device. In development: there is no Aurora account, and a session stays in the app on the phone that recorded it.
• Aurora on a Mac. In development: a macOS target exists in the app's source and there is no release, so there is nothing to install on a Mac.
• Naming a speaker yourself. Speaker names come from the conversation, inferred on the device. There is no control that types your own name over a lane.
• Tapping a word to hear it. The app has no playback, so there is no timestamp to tap and no audio to seek through.
• Exporting anything but plain text. Plain text is the only format a transcript leaves in.
• Sending a diagnostic report without your session in it. A report started inside a session opens with the switch that attaches that session's transcript, chat, recording and a screenshot already on. Turn it off before you send, and the conversation stays on the phone.

See also

1. What Aurora does — the whole capability list, with what is available, experimental and in development.
2. Try the free beta — the same TestFlight link, the installation steps and what the beta covers.
3. Privacy Notice — every network path in full, and what a diagnostic report carries.
4. How Aurora decides a live word is safe to show — the rules behind the two zones in the transcript.
5. What Aurora's speaker attribution actually scores — measured error rates for the lanes, on four suites.

Reviewed 21 September 2026 against the Aurora app source. The build changes every week; if a line here is wrong, it is a bug in this page and we want to hear about it at hello@auroraisyours.com.