Your spare iPhone is now an AI voice recorder.
Nothing leaves it.

Three ways. Hardware: they sell you a recorder; OdiMeet uses the iPhone you already have, or the voice recorder in your drawer. Where the audio goes: they upload it to their cloud and meter it by the minute; OdiMeet transcribes on the phone with no account and no meter. Whose AI: they pick the model and bill you monthly; OdiMeet uses your own Gemini key, so you pay Google's actual price and see every request.

Audio never does. Transcript text leaves only if you turn on Gemini summaries — and then it goes directly from your phone to Google with your key; there is no OdiMeet server in between. If you also enable Google Search fact-checking, specific claims in the transcript become search queries. Both are off by default and switched on per recording.

Any iPhone that runs iOS 26 — from the iPhone SE (2nd generation) and iPhone 11 up. Apple's live engine is comfortable on all of them. Whisper large-v3-turbo runs on all of them too, but on an SE-class phone it is slow and its first load takes a few minutes; the smaller Whisper model is the better fit there. A recent iPhone with an OLED screen also makes stealth mode truly black.

The first time a Whisper model loads, iOS compiles it for your phone's Neural Engine. OdiMeet does this right after the download, in Settings, with a visible timer, so your first real recording starts instantly. The result is cached; later loads take seconds.

Whisper handles code-switching within a sentence well when the transcription language is set to Chinese — English words come out in English, inside the Chinese sentence. For Taiwan-style mixed meetings that is the setting we recommend. Auto-detect listens to the first 30 seconds and picks one language, which suits imported recordings whose language you do not know.

Not yet. The next release adds multi-phone attribution described above — each participant's phone records its own track, and the loudest track wins each moment. It is the only approach we know of that does not rely on an AI guessing from voice prints, and it needs no hardware. Single-microphone speaker labels are being evaluated separately and will be labelled clearly as an estimate if they ship.

For most meetings, no. An iPhone on the table within about half a metre of its speaker records clearly, and a phone per participant is enough for speaker attribution. An external wireless mic earns its place in three situations: noisy rooms or long tables (a lavalier at the collar is 15–20 dB closer to the voice than a phone on the table, roughly twice the separation); when attribution must be near-perfect, as in depositions; and when phones can't sit on a table — walking interviews, field work. If you already own one, it plugs straight into the iPhone over USB-C, and mixing worn mics with table phones in the same meeting is fine.

Yes. A receiver plugged into USB-C is picked up as the input automatically, at 48 kHz. Transmitters that record internally (Insta360 Mic Pro, DJI Mic) show up as storage — import those files like any recorder's. Multi-transmitter sets feed the same speaker-attribution pipeline as multiple phones.

Seven days free, then $2.99 a month, $29.99 a year, or $39.99 once for lifetime — see Plans. Every plan includes everything; subscribe only for the months you need. Gemini usage is paid directly to Google with your own key, typically well under a dollar for a two-hour meeting with live summaries on.

OdiMeet is in private beta on iOS 26 now. Write to us if you record meetings, interviews or depositions and want in. App Store release is planned for 2026.