How inneedai works, from install to the moment it matters.

Before any of the setup, it is worth being precise about what the app actually does — because most of the questions we get are really questions about this.

1. 1

   It hears the other side of the call, not you

   The desktop app captures your computer's audio output — the voice coming out of your speakers. That is the interviewer, the panel, the client. Your microphone is never opened, and your own speech is never transcribed or sent anywhere.

   This is not a setting you can get wrong. There is exactly one audio source and one transcription stream, and it is the one playing through your machine.

2. 2

   It works out which parts were questions

   A live transcript is a wall of text. The app watches it for the shape of a question and marks the moment one lands, so it can answer the thing you were asked rather than the last thing that was said.

3. 3

   It answers from your material, not from the internet

   The answer is generated against the document you uploaded and the details you gave for this specific session. That is what keeps it in your own voice and your own history, instead of producing something generically true and obviously not yours.

4. 4

   You stay in control of when it speaks

   Nothing is ever said out loud, typed for you, or sent to anyone. An answer appears on your screen, in a window only you can see, and what you do with it is entirely yours.

Everything the app says on the day is built from two things you provide here: a document about you, and a short brief about the conversation. This takes about two minutes and only has to be done once per session.

1. 1

   Create your account

   Sign up with email or Google. You land on your dashboard, which is where sessions live and where you'll come back to after each call.

2. 2

   Upload your material

   Drop in the document that should ground your answers — a résumé or CV for an interview, a thesis or paper for a viva, a brief or deck for a client call. PDF, up to 10 MB. It is parsed automatically; there are no forms to fill in afterwards.

   • Everything you upload lives in your Library: two résumés and up to twenty source documents. Any session can be pointed at any of them, so a thesis is uploaded once and reused for every viva.
   • Once parsing finishes you'll see what it pulled out — the experience and skills from a CV, the title, structure and key claims from a thesis or a deck. Read that back: it is the cheapest check that the right file was read, and read whole.
3. 3

   Interviews: add supporting documents

   An interview can draw on your résumé plus up to 5 supporting documents — the job description, your portfolio, notes on the company. They have a step of their own straight after the résumé is analysed, and you can add or remove them later from the session page. PDF, Word or Excel, up to 32 MB each. Supporting documents come with the Pro and Power plans.

   • Upload only your résumé in the résumé step. Anything merged into that file is read as part of your career history.
   • Your résumé grounds every answer. Supporting documents are searched per question, and only the passages that match it best are used — so short, focused documents beat one long one.
   • Add everything before you go live, and give large files a few minutes to process. A text-based PDF reads far better than a scan or a photo.
4. 4

   Create a session for the conversation

   A session is one specific event. Give it a title, the company or institution, the role or topic, and the date and time in your own timezone. Add a description or agenda if you have one — a job posting pasted in here measurably improves what comes back.

   • Interview — a job interview, technical or behavioural.
   • Viva — a thesis defence or oral examination.
   • Sales call — answers from your deck and pricing, never an invented number.
   • Meet — a client call, pitch, or any other conversation you'd rather not walk into cold.
5. 5

   Pick a tone

   Speak as me writes the answer in the first person, as something you could say out loud more or less verbatim. Just the answers strips it back to the facts and the structure, for when you'd rather phrase it yourself. On a paid plan you can also write your own prompt instead. That switches off our tested instructions, including the rules that stop the AI making things up, so use it only if you know what you want. You can change this per session.

6. 6

   Teach it the words it will hear

   Live transcription mishears proper nouns it has never seen: a framework, a drug name, your supervisor's surname. The session's steering fields let you hand those over in advance — a short note about the conversation and a list of key terms.

   • Keep terms short and specific. Names, products, acronyms, technologies.
   • The site suggests terms from your document and job description automatically — add anything it missed.

   Worth doing for anything jargon-heavy. A viva or a specialist technical round is where it makes the most visible difference.

The website is where you prepare. The desktop app is what's actually running on the day — a small always-on-top window that listens and shows answers.

12345

1. 1

   It is listening — The dot and the clock tell you capture is running. Beside them is the balance you have left, so you never have to leave the call to check it.

2. 2

   What it heard — The question it picked out of their speech, with how confident it is that it was a question at all. If this is wrong, the answer under it will be too — which is what the key terms in your session settings are for.

3. 3

   The answer — Streamed in as it is generated, grounded in the document you uploaded. This one is in the “speak as me” tone — first person, ready to say out loud.

4. 4

   Ask your own — Ctrl+K puts the cursor here. Anything you type is answered against the same material.

5. 5

   When it misses — Regenerate takes another run at the same question. Copy lifts the text out. Clear wipes the panel before the next one.

1. 1

   Download and install

   Windows 10 and 11, 64-bit. The installer is signed and the app updates itself from then on, so this is the only time you download anything. macOS is not out yet.

2. 2

   Sign in

   The app opens your browser to sign in and hands the session straight back — you never type your password into the desktop window. If your browser asks for permission to open the app, that's the handoff, and it's expected.

3. 3

   Let it hear your computer

   On first run it needs permission to capture your system audio. Grant it once. If your meeting software is set to route audio to a headset, that still works — the app taps the output, not the speaker.

4. 4

   Learn the window

   It is deliberately small. Along the top: a status light, the elapsed time, and how many hours you have left. Below that the answer panel, and along the bottom a composer reading “Ask AI anything…” with Clear, Regenerate and Copy beside it. Drag it anywhere; it remembers where you left it.

   Put it on a second monitor if you have one, or tucked into a corner of your call window if you don't. Time spent positioning it before the call is time you don't spend fumbling during one.

5. 5

   Open the settings once

   The gear gives you opacity, text size, whether the window floats above everything, and which keys do what. These are the settings that have worked best for people, and they are worth setting before your first call rather than during it.

   • Stealth mode: on. It hides the window from screen sharing and recording, and it ships switched off. See Staying invisible.
   • Auto-answer: off, as it comes on a new install. It is still being tuned — it guesses which parts of the conversation are questions, and it can fire on a passing remark or miss a real question. With it off, nothing is answered until you press F9.
   • Answer hotkey: F9, as it comes. It works while your call app has focus.
   • Show/hide: Ctrl+Alt+Space minimises the overlay and brings it back. The window has no taskbar button and no tray icon, so this key — or opening the app again — is the way back. Change it to anything you like.
   • Meeting audio: the output device the app listens to. Leave it on System default unless your call plays through a headset that isn't the default.

   Pick an answer model tagged Ultra-fast — GPT-OSS on every plan, GPT-5.4 nano on Pro and Power. They start answering almost as soon as the question ends. The larger models write fuller answers but take longer to begin. You choose the model per session, and can switch it from the model pill on the overlay.

During a call you cannot be reaching for a mouse. The bindings below are the entire live interface, and the two that matter sit next to each other so a finger that finds one has found the other.

1234

1. 1

   Stealth mode — Hides the window from screen sharing and recording. Off until you turn it on — the one setting worth changing before your first call.

2. 2

   Auto-answer — Off on a new install: nothing is generated until you ask for it. Switch it on and a detected question answers itself.

3. 3

   The answer key — F9 out of the box, and the only binding registered with the operating system — which is what lets it fire while your call app has focus.

4. 4

   The screen key — F10, for questions about something on screen. Reset puts both back if you rebind yourself into a corner.

1. 1

   Why F9 and F10

   They collide with nothing. Zoom's shortcuts start with Alt, Teams' with Ctrl+Shift, and Meet's with Ctrl — while F1, F5, F11 and F12 are already spoken for by every browser. F9 and F10 are the two keys in the room nobody else wants.

2. 2

   You can rebind them

   Any key, with or without modifiers, in the app's settings. One thing to understand before you do: while a session is running, the app claims that key from the entire operating system.

   Bind a letter key or the space bar and that key stops reaching every other app for as long as the session runs — including the meeting chat and the one you're typing your code answer into. The app warns you when a binding has that cost. It releases the key the moment you stop the session.

Ten minutes of rehearsal with the documents you'll use on the day is the cheapest insurance there is. You find out how the app hears your setup, how quickly your model answers, and whether stealth mode is really on — while none of it matters yet.

1. 1

   Rehearse with the real material

   Use the same documents, model and tone you'll have on the day. On a paid plan you can rehearse on the real session itself: a session that has ended can go live again. On Free it can't, so rehearse on a separate session and keep the real one fresh.

   A rehearsal is billed like a real call — capture time and AI credits — so a handful of questions is plenty.

2. 2

   Call yourself from a second device

   Start a Zoom, Meet or Teams call on your computer with the app running and the session live. Join the same call from your phone or a second laptop, and ask your questions from there — the way an interviewer, examiner or client would.

3. 3

   Keep the call's sound playing on the computer

   The app hears what your computer plays, so the call has to be audible on the computer. Two devices in one room will echo; stop it by muting the computer's microphone in the call (or wearing headphones on the computer) and turning the phone's volume down. On the real call there is only one device on your side, so there is nothing to echo.

   Don't mute the call on the computer to stop the echo. It's the one fix that also silences the app — the waveform goes flat and nothing is transcribed.

4. 4

   Practise the way you'll do it for real

   With auto-answer off, let the question finish, then press F9. You don't need an answer to everything the other side says — and a press mid-sentence answers only what has been said so far.

   • What they said leading up to the question goes in with it, so a long story before the question isn't lost.
   • Want a different answer? Press F9 again; it asks you to confirm before replacing the one on screen.
   • If a question was misheard, click the overlay, press Ctrl+K and type it.
5. 5

   Check what a screen share shows

   Share your screen in the practice call and look at it on the other device. With stealth mode on, the overlay isn't there. If you can see it, fix that now rather than on the day.

This is the whole run of a call, from the click before it starts to the click after it ends.

1. 1

   Go live on the session

   On your dashboard, hit Go live on the session you're about to have. That's what tells the desktop app which conversation this is — it opens on the live session and pulls in that session's document, brief and tone. If the app is already open, it switches over by itself.

   Only one session can be live at a time. Going live on a second one quietly takes the first one offline, which is also how you fix it if you armed the wrong one.

2. 2

   Start listening

   Start the capture in the app just before the call connects. The status light goes live and the waveform starts moving as soon as it hears them. If the waveform is flat while someone is talking, it isn't hearing the call — see the last chapter.

3. 3

   Let it answer, or ask for one

   When it recognises a question it puts it at the top of the window, and the answer begins streaming underneath. When it doesn't recognise one, a question phrased as a statement, a follow-up, a pause you want to fill, press F9 and it answers what was just said. If an answer comes back at the wrong angle, Regenerate takes another run at it.

   Answering automatically is a setting, and it is on unless you turned it off. With it off, nothing is ever asked until you press the key — which some people prefer, because every answer spends time from your balance.

4. 4

   Ask your own question

   Ctrl+K puts your cursor in the composer. Type anything — a clarification, a definition, something you want framed differently — and press Enter. It answers with the same grounding in your material.

5. 5

   Answer what's on screen

   If the question is visual — a shared diagram, a snippet of code, a slide — F10 captures your screen and answers what's on it. It uses what they said too: if they show unfinished code and explain the task out loud, press F10 once they've finished, and the answer does what they asked. Available on plans that include screen capture.

   The overlay excludes itself from that capture, so an answer already on screen can never be fed back in as if it were part of the question.

6. 6

   Stop when the call ends

   Stop the capture. That releases the keyboard shortcut back to your system and stops the clock. Then conclude the session on the website when you're ready to look back at it.

Second Screen puts the question and the answer on your phone as they stream, with buttons that run the overlay's own actions. There is nothing to install on the phone: it is a page in its browser.

123456

1. 1

   Answer now — Tap it while they're still talking and it answers the moment they finish; tap again to answer right away. It is the overlay's own Answer now, run on your computer.

2. 2

   Clear — Throws away the question it's holding, so the next answer starts from what they say next.

3. 3

   Screen — Asks your computer to read your screen and answer what's on it. It appears on plans with screen capture; with several monitors, you pick one on the computer.

4. 4

   The answer, as it streams — The overlay's answer, word for word, with how fast it started and finished and which of your documents it came from.

5. 5

   Earlier answers — Every finished answer from the session, one tap away, so the live screen only ever holds the current one.

6. 6

   The link — Green and counting while the session runs. If the phone loses its connection the dot turns amber, so an old answer never passes for a new one.

1. 1

   Turn it on when you go live

   In the Go live window, tick “Also show answers on my phone”, and a QR code appears once the session is live. Already live? Use Show on phone on the session page, or the phone icon on the overlay's control strip.

   The option only appears when your plan includes Second Screen. The pricing page lists the plans that do.

2. 2

   Scan the code

   Point the phone's camera at the code and open the link. The page says it's connected and waits; when you press Play in the desktop app, the question and the answer appear as they stream. It's fastest when the phone is on the same Wi-Fi as your computer, and mobile data works too.

3. 3

   Use the buttons

   Answer now works like the overlay's: tap it while they're still speaking and it answers when they finish, or tap it again to answer at once. Clear throws away the question it's holding. Screen, on plans with screen capture, reads your screen — with several monitors, you pick one on the computer.

   • Earlier answers sit behind the count at the top of the page.
   • Aa sets the text size and shows which routes are connected: Wi-Fi, Direct or Internet.
   • Safari on iPhone can't go full screen; tap aA in the address bar, then Hide Toolbar.
4. 4

   Stop sharing when you're done

   The link ends with the session, or as soon as you press Stop sharing on the website or the overlay. Going live again makes a new code, and the old one stops working.

A tool like this is worthless if it shows up on the call. Here is exactly what protects you, and — just as importantly — what doesn't.

1. 1

   Turn on stealth mode — it ships switched off

   In the app's settings, Stealth mode marks the window as protected content at the operating-system level. With it on, the capture that leaves your machine when you share your screen simply does not contain the overlay — not blurred, not blacked out. Absent. This holds whether you share a single window or your whole display.

   Nothing else in this tutorial depends on a setting. This does. Stealth mode is off by default, so if you have not turned it on, the overlay is as visible to a screen share as any other window. Switch it on once, before your first call — it is remembered from then on.

2. 2

   It is always excluded from its own screen capture

   When you press F10 to ask about something on screen, the app switches protection on for the duration of the grab and puts your setting back afterwards. So the overlay is never in that image, whether or not you run with stealth mode on.

3. 3

   Nothing is audible, ever

   The app has no audio output. It never speaks, never plays a tone, and never touches your microphone, so there is nothing for anyone to hear.

4. 4

   It sits above your call, and stays there

   The window floats above other windows so it doesn't get buried mid-question, and you can dial down its opacity so it sits quietly in your peripheral vision instead of demanding your attention.

5. 5

   No taskbar button, no tray icon

   The window never appears in the taskbar or the system tray. Minimise it and Ctrl+Alt+Space brings it back, or open the app again from the Start menu. Like every program on the machine, it is still listed in Task Manager.

The session doesn't disappear when the call ends. What was asked, what was answered, and what it cost are all on your dashboard.

1. 1

   Conclude the session

   Marking a session done files it into your history and frees you to arm the next one. On a paid plan, a concluded session can be brought back to life later if you need it again — a second round with the same company, say. On Free, start a new one.

2. 2

   Read the recap

   Every session keeps the questions you were asked and the answers that were generated. Read back through a round to find the questions you fumbled, or to prepare for the next stage with the same people.

3. 3

   Check what it cost

   Time is billed against your balance while a capture is running, and stops the moment you stop it. You don't have to wait until afterwards to see it: the overlay's top bar shows the hours you have left, next to the clock, the whole time it is running. Your credits page has the balance, what each session consumed, and your receipts.

4. 4

   Delete anything you'd rather not keep

   Any individual session, or your whole history, can be deleted from the dashboard. It goes for good.

You are pointing this at the most consequential conversations you have. You should know precisely what is kept.

1. 1

   Audio is never retained

   Sound is turned into text as it arrives and discarded. There is no recording of your call on our servers or anywhere else, and no way to produce one after the fact.

2. 2

   Your microphone is never opened

   Only your computer's audio output is captured. Your own voice never leaves your machine, because it is never picked up in the first place.

3. 3

   Transcripts and documents are yours to delete

   The transcript of what the other side said, and the document you uploaded, are stored against your account so you can review a session later. Both can be deleted from the dashboard at any time.

Most of what goes wrong on a live call happens somewhere between the other person's microphone and your screen, in places no app controls. Knowing them in advance is most of the fix.

1. 1

   Speech recognition mishears

   Accuracy drops with strong accents, a weak microphone or a poor connection on their side, background noise, and people talking over each other. Names, companies and technical terms are what it gets wrong most.

   • Add the words you expect to hear to the session's key terms.
   • If it still struggles, set the overlay's Accuracy to Best accuracy. Text arrives a little later.
   • After the call, list the words it got wrong in the feedback form. That feeds future sessions.
2. 2

   Where a sentence ends is a guess

   The speech engine decides a turn is over from pauses and wording. A long pause can split one question in two; a quick follow-up can merge two into one. Every real-time speech model works this way — which is why pressing F9 yourself is more dependable than auto-answer.

3. 3

   It hears everything your computer plays

   Notification sounds, a video in another tab and music are transcribed along with the call. Close them and turn on Do Not Disturb before you start. Your microphone isn't used, so your own answers are never heard.

4. 4

   AI providers have slow minutes

   How fast an answer starts depends on the model's provider and on your connection, and providers occasionally slow down or fail. If answers are slow or erroring, switch models from the model pill on the overlay — ideally to one from a different provider.

5. 5

   Your network sits in the middle

   VPNs, proxies and office or school networks can slow the app down or block it. If it says the connection to the speech engine was lost, press play to reconnect.

6. 6

   Answers can be wrong

   A model can misread a question or get a detail wrong, so read the answer before you say it. It is told not to invent details your documents don't contain — so the figures, dates and tools you want to quote need to be in them.

7. 7

   Your balance can run out mid-call

   If your AI credits run out, the transcript keeps going but answers stop until you top up. If your time runs out, the session ends. Check both before the call, not during it.

In rough order of how often we see them. If none of these is it, contact support — and say what the status light and waveform were doing, because that narrows it immediately.