For iPhone, iPad and Mac

Live translation, transcripts
and answers while you talk.

GPT Realtime Assistant is a live AI assistant for iPhone, iPad and Mac. Start a session, and it hears what is said, writes it down, and answers the question on the table before the silence gets awkward: a fact, a phrasing, the next step. Or it interprets a conversation both ways: each person speaks their own language, and every sentence appears, and is read aloud, in the other. Afterwards the session is a record with tasks and notes attached. You press Start and End; on iPhone and iPad, keep the app open while the session runs.

Subscription. AI credits included. Not affiliated with OpenAI. · iOS 18+ · iPadOS 18+ · macOS 26+ (M1+)

Interactive preview · Demo01 · Live Voice & Transcript OpenAI Realtime API · Live Voice
GPT Realtime Assistant on iPhone: the Live screen with a transcript, an assistant answer and the credit balance.

In the moment · Short films

When the answer can't wait.

Conversations, classes, travel and everyday surprises rarely follow a script. See how GPT Realtime Assistant can help you make sense of the moment.

New · 19 seconds · Two-way translation

Stranded in Tokyo. Words in the way.

A traveller and a mechanic try to bridge the language gap. Watch a recorded English–Japanese translation example with GPT Realtime Assistant.

AI-generated scene; app footage is from a recorded test, edited for time. Translations can be imperfect: check important details.

Watch on Instagram

New · 11 seconds · Photo context

A $12 surprise on the bill.

A restaurant bill can raise questions. Attach a photo and ask GPT Realtime Assistant to explain a sample service charge.

AI-generated scene and voice; illustrative receipt. The answer is from a real app test; the iPhone view is composited and the sequence is edited for time.

Watch the Short

11 seconds · Interview questions

When the next question comes fast.

What happens when an interview takes an unexpected turn? This playful short imagines GPT Realtime Assistant helping with a technical question.

The scene and voices are AI-generated and scripted; the app screen is a composited example.

Watch the Short

42 seconds · Voice, notes, answers

Stay with the conversation.

Follow a family exchange across languages, a lecture point turned into a task, and help shaping an interview response.

Scenes and voices are AI-generated; some example app text was adapted for the demonstration.

Watch on YouTube

21 seconds · Photo context

One question, a lot of moving parts.

A photo can give the assistant context. Watch it work through a sample timetable with connections, times and platforms.

The scene is AI-generated and the timetable is illustrative. The app response comes from a real test, edited for time.

Watch the Short
macOS · 1280 × 720Live translation, transcript & answers on Mac
Read tutorial →
Recorded example · JA → FRSetup, spoken French reply & reversing From/To
  • Real-time voice over OpenAI's Realtime API
  • Seventeen session templates, from interview to standup
  • Two-way voice translation between two people
  • Karaoke-style readback with word highlighting
  • Projects, sessions, tasks, scheduler
  • Offline Whisper and Qwen models on Mac

What it does

Live, then a record.

Five tabs: Live, Projects, Sessions, Tasks and Scheduler, with Notes on Mac. On iPhone, Scheduler is under More. The Live tab is where the conversation happens; the rest is what it leaves behind.

A live session, not a chat box

01 · Realtime Assistant

A live session, not a chat box

Tap Start and the microphone streams to the model. What is said appears as a transcript; what the assistant has to say appears beside it. Semantic voice detection knows when a speaker has finished, and you can interrupt the answer by talking. Answers come in brief, balanced or detailed length, and can be read aloud too.

  • OpenAI Realtime API
  • Transcript and answers side by side
  • Interrupt by speaking
  • Brief, balanced or detailed answers
Templates for the room you are in

02 · Realtime Assistant

Templates for the room you are in

A new session starts from one of seventeen templates, each with roles and instructions already written: interview, negotiation, sprint planning, daily standup, client discovery, sales demo, support triage, incident review, design review, user research, lecture notes, one-on-one, hiring debrief, decision meeting, translation, coding task and product review. Or start a blank session and write the role and instructions yourself; either way, you can edit them for the session in front of you.

  • Seventeen templates with roles and prompts
  • Blank sessions with your own instructions
  • Projects hold sessions; sessions hold history
Two-way translation and readback

03 · Realtime Assistant

Two-way translation and readback

Turn on Translate both ways, choose the two languages, and the app interprets between two people: each speaks in turn in their own language, and every sentence appears in the other language, read aloud when Speak translations is on. It runs two translations at once, in 58 languages, for about 85 credits a minute. For one direction only, a translation session puts whatever is said, in a language it detects or one you set, into the language you choose. Karaoke readback uses Apple's on-device speech recognition to highlight words as you read them, for rehearsing a talk or practising a language.

  • Translate both ways between two people
  • Translations read aloud as they appear
  • One way, with the spoken language detected
  • On-device readback highlighting
Context you attach

04 · Realtime Assistant

Context you attach

Add a photo or a screenshot to a session and the assistant answers with it in view. On Mac the app can capture a chosen screen or window, record the session, and watch system audio, so a screen-shared meeting is context too.

  • Photo, camera and screenshot attachments
  • Screen and window capture on Mac
  • System audio on Mac
Tasks, notes, scheduler

05 · Realtime Assistant

Tasks, notes, scheduler

Useful moments become tasks with status, priority, dates and comments, and evidence attached as screenshots or video. Sessions can be scheduled with reminders. Notes live alongside on Mac. All of it is stored on the device.

  • Task board with evidence
  • Scheduler with reminders
  • Notes on Mac
  • Stored locally
Local models on Mac

06 · Realtime Assistant

Local models on Mac

On an Apple silicon Mac the app can download Whisper Large v3 Turbo for transcription and Qwen 3 8B for answers and run them on the machine, for sessions whose audio should not leave the room. Local sessions work without the cloud and use no credits; their answers appear as text. On iPhone and iPad, transcription is either OpenAI's or Apple's on-device speech recognition.

  • Whisper and Qwen, downloaded once, run locally
  • Mac with Apple silicon only
  • Apple Speech on every device

Who it is for

For people who are expected to have the answer during the call.

Engineers and consultants
The technical question you half remember, answered while the other person is still finishing the sentence.
Managers
Standups, one-on-ones, hiring debriefs: a transcript, the decisions, the tasks, without a notetaker.
Learners and speakers
A lecture transcribed, a language interpreted live, a talk rehearsed with the words lighting up as you say them.
Not for
Rooms where assistance or recording is not allowed. You are responsible for permission from the people you are talking to.

What it costs

Monthly or yearly access. AI credits included.

The app needs an active subscription to run: $9.99/month (2,000 AI credits each month) or $79.99/year (48,000 AI credits per annual payment — equivalent to 4,000 credits/month, 2× the monthly credit rate while saving 33% vs monthly billing). A thousand credits represent one US dollar of model usage at OpenAI's list rates, metered by the audio and text the session actually uses. Optional 1,000- and 5,000-credit packs add more without changing your subscription. We plan to change the US yearly price to $119.99 and grant 2,500 AI credits each month for 12 months, for a total of 30,000 credits.

Download

$0 free to install

  • iPhone, iPad and Mac
  • Subscription across supported devices
  • Restore purchases on any device

The app needs an active subscription to run.

Yearly

$79.99 per year

48,000 AI credits/yr · 2× monthly rate · Save 33%

  • Full app on iPhone, iPad and Mac
  • 48,000 AI credits per annual payment (4,000/mo equivalent)
  • All templates, translation, readback
  • Tasks, scheduler, notes & local Mac models

Billed annually upfront. Renews yearly; cancel anytime in your Apple account.

Monthly

$9.99 per month

  • Full app on every device
  • 2,000 AI credits each month
  • All templates, translation, readback
  • Tasks, scheduler, notes
  • Local models on Mac

Renews monthly. Cancel anytime in your Apple account.

1,000 credits

$4.99 one time

  • About one US dollar of model usage
  • Active subscription required

A consumable pack, added to your balance.

5,000 credits

$19.99 one time

  • About five US dollars of model usage
  • Active subscription required

A consumable pack, added to your balance.

Credits are spent on what the session uses: audio in and out costs more than text, cached context costs less. Two-way translation is billed by the minute of audio, about 85 credits a minute. A session reserves a small amount to start and refunds what it did not use when it ends.

Pricing →

Questions

Asked often enough to write down.

Can it translate a conversation between two people both ways?

Yes. Turn on Translate both ways and choose the two languages; 58 languages work in both directions, without Auto Detect. Each person speaks in turn in their own language, and every sentence appears in the other language and, with Speak translations on, is read aloud. It runs two translations at once and uses about 85 credits a minute, so the 2,000 monthly credits cover roughly 23 minutes of it (and the 48,000 yearly credits cover roughly 564 minutes across the year).

Is this made by OpenAI?

No. GPT Realtime Assistant is made by Arte Soft and uses OpenAI's Realtime API as its model provider. It is not affiliated with, sponsored by or endorsed by OpenAI.

Do I need my own API key?

No. The subscription covers model usage through the app's own backend, which issues a short-lived session secret so your device can talk to the model directly. There is nothing to configure.

Where does my audio go?

During a session, from your device straight to OpenAI's Realtime API, under OpenAI's API data policies. Our backend never receives the audio. It records usage, purchases, credits and the session's title so it can bill and reserve credits; it stores no transcripts and no audio.

What happens to the transcript?

It stays on your device with the session. There is no cloud sync; sessions on your Mac are not on your iPhone.

How far do 2,000 monthly or 48,000 yearly credits go?

It depends on how much of the session is speech: audio costs more than text, and brief answers cost less than detailed ones. The yearly plan grants 48,000 AI credits per annual payment (4,000/month equivalent — twice the monthly plan's 2,000 credits). The balance is on screen during every session, and local sessions on a Mac use no credits.

Can it work offline?

On an Apple silicon Mac, with the local Whisper and Qwen models downloaded, transcription and answers can run without the cloud, and use no credits. On iPhone and iPad, sessions need a connection; only Apple's speech recognition runs on the device.

Is the assistant always right?

No. Answers can be incomplete, wrong, late or beside the point. Treat them as a prompt to think, not as advice, and check anything that matters.

Can I record other people?

Only with their permission and where it is allowed. Recording and assistance rules differ by workplace, school, jurisdiction and interview; you are responsible for following them.

Start the next session with backup.

On the App Store for iPhone, iPad and Mac. Subscribe, pick a template, press Start.

Free Voice Resources

12 Zero-Upload Browser Voice Tools & 140 AI App Comparisons

Check your microphone level and room echo before a call, calculate OpenAI Realtime token costs, time a 90-second STAR interview answer, or see how GPT Realtime Assistant compares against 140 voice, meeting, and translation tools.

Mic Level & Echo Test → Live dBFS meter + 5-second local playback check. Realtime Voice Cost Calculator → Compare API token bills vs $79.99/yr + $0 offline Mac. Speaking Pace (WPM) Analyzer → Check WPM speed and spot filler words instantly. Mac Whisper & Qwen RAM Check → Verify unified memory sizing for M1–M4 offline AI.
Explore all 12 Free Voice Tools → Compare vs 140 Voice & Meeting Apps → vs Wispr Flow vs Granola AI vs tl;dv Best Voice AI Apps (2026)