Every question, including the uncomfortable ones

It listens to a live conversation — your microphone and everything coming out of your speakers — transcribes it as it happens, and writes your next move before you have to think of it. In a technical interview that means an approach, working code and its complexity, and calls. In a presentation it means talking points and answers to the hostile question you are about to get.

About five minutes. Install the app, download one speech model (40 MB on average), paste an AI provider key, and pick your microphone and output device. There is no account to create and nothing to configure on the other side.

No. CoderCue has no sign-up, no email capture and no login. Your licence key is the only credential, and it activates the app on this machine.

Windows 10 and Windows 11 are fully supported today. The capture and invisibility layers are built on Windows APIs, so macOS and Linux are not available yet.

Yes — no card, no trial countdown. You get five minutes of live assistance per day, resetting at midnight, and you keep every session transcript you have already recorded.

No. The coding playbooks are the sharpest ones, but the same engine handles system design, investor meetings, negotiations, lectures and interviews. Switching playbook changes the entire personality of the assistant.

CoderCue captures two independent streams: your microphone, tagged YOU, and the system output, tagged INTERVIEWER. Attribution comes from which stream the audio arrived on, not from voice recognition, so it is reliable from the first second of a session.

Vosk — an offline, on-device recogniser. Nothing you say is uploaded to a transcription service, and a dropped internet connection does not stop transcription. Nineteen models covering eighteen languages are available for download, including a 128 MB high-accuracy English model.

Transcription does, because the model runs locally. Answers need an internet connection to reach your AI provider — if the connection drops mid-session CoderCue keeps transcribing and queues your question so nothing is lost.

Yes. CoderCue captures every active playback device on the machine, not just the Windows default. That matters when Discord, Zoom or a browser are routed to a headset while Windows still thinks the default output is a monitor or a USB speaker.

Yes, and you should not have to shout. The sensitivity slider sets the level at which speech starts being captured, and the noise gate adapts to your room automatically. If your microphone is quiet, turn on input boost in Settings — it normalises level in real time and is off by default because it also lifts room noise.

That is the room, not you — a fan, a keyboard, or a TV. The gate learns your room's noise floor over the first few seconds and lifts its threshold above it. Lower the sensitivity slider if it is still triggering, and check that the correct microphone is selected rather than a monitor or loopback endpoint.

Yes. Download the model for your language in Settings and switch to it — Spanish, French, German, Italian, Portuguese, Dutch, Polish, Russian, Ukrainian, Turkish, Hindi, Mandarin, Japanese, Korean, Vietnamese, Persian and Czech are all available.

A playbook is the persona and output contract behind every answer. There are eleven built in: Algorithm Coach, System Design, Complexity Radar, Bug Hunter, Take-Home Review and Refute, Presentation Copilot, Meeting Copilot, Negotiation Coach, Live Summarizer, Delivery Coach and Blank Slate. You can also write your own.

Any OpenAI-compatible chat API. CoderCue is built and tuned for DeepSeek, whose models are fast, cheap and strong at argumentation — but the endpoint and model name are both editable, so you can point it at a different compatible provider.

Because it is cheaper for you and more honest. You pay your provider at cost — typically a few cents per session — instead of a markup hidden inside a subscription. Your key is stored encrypted on your machine and never sent anywhere except your provider.

Yes. Add PDFs, text files, Markdown or images to the knowledge base and CoderCue uses them as its primary source of truth, falling back to the model's own knowledge only when your material does not cover the question. Images — slides, screenshots, whiteboards, handwritten cards — are read by a vision model and stored as text, so they only cost tokens once.

Yes. The vision model accepts JPEG, PNG, GIF and WebP images up to roughly 1024 tokens each. Drop the image in and the extracted text becomes part of your knowledge base.

No. CoderCue builds your knowledge context once per playbook and relies on provider-side prompt caching, where a cache hit costs around one fiftieth of a cache miss. Your documents sit in the cached prefix, so the tenth answer in a session costs a fraction of the first.

It is explicitly instructed not to. Every built-in playbook forbids inventing studies, cases or quotations, and where a fact cannot be verified the answer says VERIFY and describes what to check instead of producing a confident fake number.

Yes, on Windows, with the Invisible plan. CoderCue asks Windows to exclude its window from capture using the same display-affinity API that DRM-protected players use. The window is completely normal to you and absent from screen recordings, screen shares, Discord streams and OBS output on the same machine.

No — nothing can be. Invisibility applies to software capture on the same machine. A phone camera aimed at your monitor sees your monitor.

No. There is no CoderCue backend in the audio or answer path. Audio is transcribed on your machine and only the text of the conversation goes to your AI provider, over your key. Your transcripts, knowledge base and settings stay in local files you can delete at any time.

It stores sessions, settings and downloaded models in your user app-data folder. Nothing is added to startup by default, no background service is installed and no telemetry is sent.

That depends entirely on the rules of your league or event, and you are responsible for knowing them. Many competitive circuits prohibit outside assistance during a round. CoderCue is a tool for practice, preparation, coaching, and formats that permit assistance — use it accordingly.

Standard is unlimited live assistance with everything except the invisibility layer. Invisible adds screen-capture exclusion, so the window cannot be seen in a recording or a screen share on the same machine. Both include every playbook, the knowledge base, unlimited sessions and unlimited history.

No. You buy a licence for a fixed term — one month at a time — with no recurring billing and no card kept on file. When the term runs out the app drops to the free plan and you buy another key if you want to carry on. Nothing renews behind your back.

Your key arrives by email the moment payment clears. Open CoderCue, go to Settings, paste the key and press Activate. The first activation binds the licence to that machine.

A licence activates on one machine at a time. If you replace or rebuild your computer you can move the licence to the new machine yourself from the licence page — the previous machine is released automatically.

Nothing. CoderCue caches your entitlement locally and keeps working for a grace period without reaching the licence server, because an interview should never stop because the venue wifi dropped.

No — licences are sold as final sales. The free plan exists so you can decide before paying rather than after, and anything we get wrong — a duplicate charge, a key that never arrived, a key that will not activate — is put right in full. Rights you have under your own country's consumer law are unaffected.

The compact English model runs comfortably on a modern laptop alongside the rest of your work. The high-accuracy model is heavier; if you are on battery, use the compact one.

The app ships without a speech model so the installer stays small. You download only the model for the language you actually speak — 40 MB for compact English, 128 MB for the accurate one.

Yes. Every session is saved with its full transcript, speaker labels and AI answers, and can be read back, searched, or copied out as Markdown for a brief, a file note or a coaching review.

Yes. If Windows exposes the headset as an active playback device, CoderCue captures it. Note that Bluetooth headsets often switch to a low-quality hands-free profile the moment an app opens the microphone — if your audio quality drops, disable the headset's handsfree device in Windows sound settings.