PRODUCT · 8 MIN READ
An AI note-taker without a meeting bot: why Aurora records from the room
Put your iPhone in the room and record through its microphone. Nobody joins your call, and four things are harder because of it, including two setups Aurora simply cannot hear.
Published · Updated
Do I need a meeting bot to get AI meeting notes?
No, not if you are in the room. Aurora records through the microphone of your iPhone, so nothing joins the call, and it works on conversations with no calendar entry. It suits in-person meetings, interviews and lectures; a call on a laptop speaker beside the phone works, with worse audio. It cannot hear a call playing in your headphones, and it cannot record a meeting you skipped: for those, a meeting bot that joins the call is the right tool. As of October 2026, Aurora runs on iPhone only; the Mac app is not released.
Put your iPhone in the room and record through its microphone. Aurora does not join the call as another participant.
Most AI notetakers begin with a calendar invitation. You connect a calendar, the service joins your call as an extra participant, and a tile appears in the grid with a product name on it. Aurora does not work that way, which buys a great deal, and costs some things this article also lists.
What does a meeting bot actually do?
A meeting bot is a server-side client that authenticates against your conferencing platform and joins the call like a person would. Once admitted it receives the mixed audio stream, and often the video and the chat, then uploads that stream to the vendor's infrastructure for transcription and summarization.
The consequence people notice is social: everyone in the room can see that a recorder from a company they have no relationship with has entered the conversation. The consequence they do not notice is architectural. A bot-based notetaker cannot process locally even in principle, because the audio arrives at a server before it arrives anywhere else. Privacy in that design is a policy commitment about how the recording is handled after collection, not a property of where the recording lives.
How does Aurora capture audio instead?
Aurora records from the device in front of you: the microphone of the iPhone in the room. There is no server-side client, no platform authentication, and no additional attendee. From the conferencing platform's point of view nothing has been added to the call, because nothing has.
Concretely: the capture path asks iOS for the record category in measurement mode, ducking other audio. That is a microphone, configured to measure rather than to flatter, with no voice processing shaping the signal before the models see it.
This is why Aurora works on conversations that have no calendar entry at all: an interview at a table, a lecture, a workshop, a hallway decision, an appointment. A bot requires a meeting to join. Aurora requires only that you are present.
A Mac app exists in the source tree and is not released, so nothing in this article describes recording on a Mac. When it ships, it gets its own paragraph.
Which setups actually work?
Recording acoustically has a hard boundary: Aurora hears what the room hears. That makes some setups good, one marginal, and one simply not a thing Aurora can do.
| Setup | Works? | Why |
|---|---|---|
| iPhone on the table, people in the room | Yes, the designed case | Everyone's voice reaches the microphone directly |
| Remote participants on a laptop speaker, iPhone beside it | Yes, with worse audio | The far end arrives as sound in the room, with the speaker's coloration and the room's reverberation on top |
| You wearing headphones or AirPods during a call | No, for the far end | The other side is playing into your ears, not into the room. The capture path does not request a Bluetooth input route, and the far end of a call is not audio iOS hands to a third-party app. You would record only yourself |
| A call happening on the same iPhone that is recording | No | iOS does not give an app the other party's audio from a phone or VoIP call. On speakerphone you are back to the acoustic case: Aurora hears the speaker in the room, not the call's stream |
Aurora's product contract does not enumerate supported capture configurations, so the rows above are read from the capture code and from what iOS permits, not from a promise. If you need the far side of a call and cannot put it on a speaker, device capture is the wrong tool and we would rather say so.
A bot-based notetaker requires you to trust a policy. A device-based recorder requires you to trust an architecture. Only one of those can be inspected.
Why Aurora captures locallyWhat does device capture give up?
Device capture is not free, and Aurora is worse than a bot at four specific things:
- Attendance without you. A bot can record a meeting you skipped. Aurora cannot: if you are not there, there is no recording.
- Per-participant audio streams. A bot admitted to a platform can often receive a separate stream per speaker, which makes attributing speech near-trivial. Aurora receives one mixed acoustic signal and has to separate speakers from it, which is a materially harder problem.
- Participant names for free. A bot inherits the roster, so it knows the call contains Priya and Tomas. Aurora separates speakers automatically but has no roster to read: a name only appears when someone says who they are in the recording and a model on the device reads it out of the transcript. There is no way to type one in, and an unnamed lane stays Speaker 1. What is found is stored with the session, not learned as a voice: a name does not carry into the next meeting, and recognizing a known voice again is still in development.
- Room acoustics. A bot receives clean per-endpoint audio. A phone on a table receives the room: distance, reverberation, an air conditioner, and the person who turned away mid-sentence.
Two of these are permanent consequences of the architecture. Two are engineering problems Aurora is actively working on, and speaker separation quality is where most of that work goes.
Does Aurora work with Zoom, Teams and Google Meet?
Aurora records the conversation rather than integrating with the platform, so the platform is not something Aurora needs to support. A call on any conferencing tool can be captured the same way an in-person conversation is captured, because Aurora is listening to the same thing you are.
The practical trade is audio quality. Capturing a call through a laptop speaker and a phone microphone is worse than receiving the platform's own stream, and it will show in the transcript. Aurora is at its best on conversations where the sound reaches you directly.
Does not having a bot make recording more discreet?
It does, and that is a responsibility rather than a feature. Removing the tile that announces a recorder removes the automatic disclosure that a bot provides, which means disclosure becomes something you have to do yourself.
Recording and consent requirements vary by country and sometimes by region, and the organization or venue you are in may have its own rule on top of the law. Before recording, ask the people present (a question, not an announcement) and find out what governs the room. Aurora gives you control over where conversation data is processed and stored. It does not remove your responsibility to the people in the conversation, and no architecture can. Recording & Consent is our practical guidance on that, and it is guidance rather than legal advice.
Related
- What on-device transcription costs on an iPhone: the compute budget that makes local capture viable.
- Privacy Notice: what stays on your device: the path from audio to a local session, and what can leave.
- Recording & Consent: asking the room, and the rules that are not the law.
- Aurora vs Granola: two note-takers with no bot, and where the audio goes in each.
