Straight answers about inneedai.

No. inneedai runs four kinds of session — job interviews, viva examinations, sales calls and client meetings — and each one is grounded in different material. An interview answers from your résumé and the role you pasted in; a viva answers from your own thesis and says which part it drew on; a sales call answers from your deck and pricing; a client meeting answers from the brief you uploaded. The framing changes with the session type, so a viva answer argues and a sales answer sells.

A Windows desktop app. It listens to the audio already playing through your machine, watches the transcript for the shape of a question, and puts an answer on screen — drawn from the document you uploaded rather than from the open internet. It is not a browser extension, and it does not join your call as a participant, so there is no bot in the attendee list.

Never. Nothing is spoken aloud, nothing is typed into your call, and nothing is sent to anyone. The answer appears on your screen and you decide what to do with it.

From the material you gave it. Uploads are parsed and indexed, and each answer is generated against the parts relevant to the question that was just asked. That is the whole design: on a sales call it will quote your own pricing rather than invent a number, and in a viva it will point at the section of your thesis it argued from.

Your computer's audio output — the other side of the call. Your microphone is never opened, so what you say is not transcribed and not sent anywhere.

By default, one key. A new install starts with auto-answer off: once a question has been asked, press F9 and it answers what was just said — which also covers questions phrased as statements. Switch auto-answer on in settings and it labels each question it recognises and starts answering by itself. Ctrl+K puts your cursor in a composer if you would rather ask for something specific.

• F9 — answer what was just said.
• F10 — answer a question that is on screen rather than spoken.
• Ctrl+K — type your own prompt against the same material.

On the fastest models an answer starts on screen within about a second of the hotkey; every model in the picker is labelled with its speed, and the deepest ones take a few seconds longer. Transcription runs continuously in the background, so the model is not starting from silence when you ask.

Yes, and with anything else that plays audio through your machine — it captures the system output rather than integrating with a specific platform, so there is nothing to install into your meeting tool and nothing for it to detect. The answer hotkeys are F9 and F10 precisely because they collide with none of those apps' shortcuts.

Yes, with Second Screen. Scan a QR code when you go live and the question and the answer stream to your phone's browser — no app to install. The phone also works as a remote: Answer now and Clear run on your computer, and so does Screen on plans with screen capture. The Windows app still does the listening.

Nothing joins the call, so there is no bot in the attendee list and no participant to notice. What they can see is your screen, if you share it. An optional stealth mode marks the overlay as protected content at the operating-system level so a screen share does not contain it — but it ships switched off, and turning it on is your decision.

No. We never use your uploads to train AI models; they are used to answer your own sessions. The AI providers that process them work under their own terms: the paid APIs we use from OpenAI, Anthropic, Google and Groq don't train on that data, while DeepSeek's terms don't rule it out — so choose another model for anything sensitive. The privacy policy names every third party involved in processing and what each one does.

Transcripts and the answers generated alongside them are kept against the session so you can read them back afterwards. You can delete a session, and deleting it takes its transcript with it.

That depends entirely on the room you are in, and it is your call, not ours. Plenty of what it is used for carries no such question at all — a sales call, a client meeting, a rehearsal. Where an examiner or an employer has set rules, those rules are the ones that count, and we would rather say so plainly than pretend the question does not exist.

The desktop app runs on Windows 10 and 11, 64-bit. A macOS build is in development and the download button for it is disabled until it ships.

You choose, from models by five providers. Free includes three fast models on Groq; holding credit adds OpenAI and DeepSeek models; Pro adds Google Gemini; Power adds Anthropic's Claude and the largest OpenAI model. Before a session, each model shows how quickly it starts answering, whether it can read your screen, whether it thinks first and roughly what an answer costs. Or leave it on your plan's default and we pick for you.

Eighteen, and it works out which one is being spoken on its own — there is no language setting. It also keeps up when a call switches between two of them mid-sentence, which is the case that usually breaks live transcription.

• English, Spanish, French, German, Italian, Portuguese, Arabic, Danish, Dutch, Finnish, Hebrew, Hindi, Japanese, Mandarin, Norwegian, Swedish, Turkish and Vietnamese.

About two minutes. Install the app, sign in, upload the document that grounds the session, and add the role, topic or agenda. The tutorial walks through it screen by screen.

From two allowances that come with your plan: hours of live-session time, and AI credits. Time counts only while the app is actually capturing audio — a session left open while you make coffee is not billed — and answers and document processing draw on the credits. Past either allowance, usage comes out of your credit balance. The pricing page lists what each plan includes.

You can buy a $5 credit pack and keep going — the pricing page shows about how much live-session time a pack buys — and credits do not expire. If you keep running out, a plan costs less per hour.

Yes, any time, from your dashboard. The refund policy sets out where a refund applies.