This is how Aurora works.
Not a video. An interactive replica of the iPhone app, in your browser. Press record and watch the transcript arrive, speaker by speaker.
What you leave a conversation with.
Follow the conversation
Speech becomes text as it is spoken: punctuated, timed and readable while the meeting is still going. Nothing waits for an upload.
When a live line is safe to readSee who said what
Aurora tells the people in the room apart and gives each one a lane. Add names to the speakers in a conversation and the transcript reads back in their names.
Names stay with that conversation. Aurora does not yet recognize the same voice in the next one.
How often it gets a speaker wrongLeave with a summary and next steps
A language model on the device writes the summary, the action items and the notes. Ask what was decided and the answer cites the transcript it came from, or says the evidence is not there.
Where the audio goes, stage by stageConversations ask for your attention. Notes ask for it too. Aurora is built to help you listen now and return to the details later, while keeping the core work on the device you already carry.
How Aurora is builtWhat stays, and what can leave.
We won't tell you that nothing ever leaves your phone. Here is the exact list, so you can hold us to it.
| What it is | Where it lives | What it takes to leave |
|---|---|---|
| The recording | On the phone that heard it. | Nothing sends it. Sharing a file is yours to do. |
| Transcript, speakers, summary and notes | Written on the device, by models on its own chip. | Nothing. No step in that path uploads them. |
| Your library | In the app on that device. No Aurora account. | Nothing. Export is plain text, where you send it. |
| Speech and language models | Installed once, then used offline. | A download at setup and when a model changes. Wi-Fi unless you allow cellular. |
| Your next calendar event | Read on the device, through the system calendar. | Nothing. Aurora sends nothing anywhere to show it. |
| Your remaining free minutes | Counted on the device, and mirrored to your own iCloud. | Automatically. This is the one always-on path, and the iCloud sync setting below does not cover it. It carries a counter, and no conversation content. |
| Setup measurements | Collected while the models install. | Only if you agree to send them. No transcript, audio or summary. |
| Selected settings | On the device. | Only with iCloud sync on, which is off by default. Settings only, never audio or transcripts. |
| A web search from Ask Aurora | Off by default. | Only if you turn it on: a minimized query, without transcript or audio. |
| A diagnostic report | Built and encrypted on the device. | Only when you send one — and the form opens with Include session details already on. That one switch attaches this conversation's transcript, Ask Aurora chat, recording and a screenshot together. Turn it off for a report with no conversation content. |
The Privacy Notice carries the same list in full, with what each service receives.
What the beta actually is.
Aurora is in TestFlight and changes every week. This is where it stands today.
What it runs on, and what setup needs
- Aurora is built for iOS 18 and newer. The Mac app is not released.
- The phone we measure on is an iPhone 15 Pro, and there is no tested-device table beyond it yet — what you need to run it carries the current requirements.
- Install through TestFlight, then download the models over Wi-Fi: a speech model of 75 MB to 600 MB and a language model of 2.4 GB or 3.4 GB. Cellular needs a setting you turn on. After that it works offline.
Available in beta
- Recording, live transcript and speaker lanes on iPhone
- Summaries, action items and notes on device
- Ask Aurora, grounded in the conversation you are in
- Offline once the models are installed
- Plain-text export
Experimental in beta
- Languages other than English: the speech model covers many, and we have no figures per language
- The second pass that refines a finished transcript
- Sessions much longer than an hour: no limit is set, and none is measured
In development
- The Mac app
- Recognizing the same speaker in a later conversation
- Search and answers across conversations
- Conversations syncing between devices
Reviewed 18 September 2026 against the Aurora app source.
Questions.
No. Transcription runs on your iPhone. You can separately choose to export a recording or include content in a diagnostic report.
Aurora is built for iOS 18 and newer. The phone we measure on is an iPhone 15 Pro, and we have no tested-device table beyond it yet. The Mac app is not released.
No. Aurora listens through the microphone of the device you are holding, in the room or next to the laptop.
Nothing. Pricing comes later, and testers hear about it first.
