Call Sessions
Prepare for calls and review past sessions. Start a session, answer live, and read every transcript back afterwards.
Call Sessions
Prepare for calls and review past sessions. Every session is free — no session credits, no upgrade.
Mock = the AI interviewer asks you one question at a time and shows the suggested answer from your documents after you answer. Interview / Regular = a live call where the interviewer's questions are answered as they are heard.
- Click the tune / lock icon at the left of the address bar (next to
copilot.appsnation.co). - Click Site settings.
- Find Microphone and change it to Allow.
- Come back to this tab, reload the page, then press Start recording — Allow mic again.
You can also open Chrome Settings → Privacy and security → Site settings → Microphone, and Allow copilot.appsnation.co there. Until you Allow it, use “Type a manual message” below — typing always works.
Press “Share Meet tab” below and choose the Meet tab.
Answers are built only from your attached documents, brief and the script added to this session — the script is used first when it matches the question. Every answer is one short paragraph in simple first-person words, plus at most 2–3 short bullets, written as natural spoken words — never lines pasted from your documents. Nothing is invented: with no document attached, the answer panel says so instead of making one up.
Documents
Add documents for more relevant AI answers. Answers in Live Session and Mock are built only from these documents and your brief.
Paste a document as text
Live Copilot
Ask, speak, or let it listen. The opening lines appear first — start speaking from them while the rest of the script streams in. If the question changes, the answer switches instantly.
Meeting simulation — screen-share auto-hide
Simulates a meeting app on this page only — a web page cannot see your real Zoom / Teams / Meet share. Pick the app, pick what you share, then start the simulated share: both answer popups hide and Safe View (plain notes) takes over. Window / tab sharing keeps the popup out of that window or tab in a real call too. Entire-screen sharing is different: everything on that display is captured — the only hiding guarantee is reading answers on your phone or a second screen.
Reading key — dark text: say it as written, it comes only from your own material. Orange bold [CHANGE: …]: the part you adapt for this professor / lab / company / question — put the real detail there before you say it.
Paste an AI reply — free ChatGPT / Claude / Muse
Using a free chat? Tap “Ask … free” in AI Connections, paste the question there, then paste the reply here and it becomes your SAY THIS answer. No API key, no payment.
Headphones? Then the mic CANNOT hear the interviewer — read this first
Plain truth: a web page's microphone only hears sound that reaches it out loud. If the interviewer's voice is playing only inside your headphones or earbuds, their voice never reaches the microphone — the mic will show sound from you, but it will never pick up their question, no matter how good the mic is. That is not a mic fault. Pick how their question will reach this page:
Speaker mode is ON: take the headphones off, play the meeting out loud on the laptop speakers, keep this device near the speakers, press “Start listening” — the mic hears the interviewer and answers on its own. (Your own mic in the meeting may echo; use the meeting app's echo cancellation, or mute yourself while they talk.)
Nothing heard yet. When the mic is working, the level bar moves with the sound in the room and the words it hears appear here live. The mic hears the meeting only if the call audio reaches it — play the call on your speakers (no headset), or put the phone / laptop speaker near this device's mic. If the bar does not move, no sound is reaching the mic and no question can be answered; if the bar moves but no words appear, speech recognition is not transcribing.
Live controls
Shortcut: press Ctrl/⌘ + Shift + H any time to hide or show answers instantly.
Transcript
Everything heard and answered, in order. This becomes your Notes & Summary after the call.
Meeting Notes — Weekly Review
- Agenda items reviewed and owners confirmed for this week.
- Open items carried forward from the last review.
- Next review scheduled; minutes to be circulated afterwards.
Safe View is on — answers are hidden. Press the button again (or Ctrl/⌘ + Shift + H) to bring them back.
Brief & Documents
The brief tells the AI what the call is about. Prepared material is the first place answers come from — lines marked SAY are spoken, lines marked DON'T SAY or NOTE stay private.
Auto script from your attached documents
Attach documents to this session (in Attached Documents below, or in Mock Test), or paste material, then press this button. Every session keeps its own separate files, and before the script is built you are asked which of this session's attached documents the script should be built from. If this session is an interview, the script is written in interview style — professor / lab wording — and each other meeting type gets its own style. It builds a numbered script in the exact format the live answers follow — 1. PRICING — ONLY IF IT COMES UP
▶ SAY
"The rebuild is 1,200, and SEO after that is 300 a month, no lock-in."
NOTE - DON'T SAY
My floor is 1,000. Only offer it if they push twice. — using only real facts parsed from your own attached documents: a section appears only when your attached documents actually have a fact for it (no pricing section without a real price in your material), NOTE lines stay private and are never spoken, and pressing Generate again rebuilds the same script from the same documents instead of degrading it. It also generates your 20-question mock bank from that script. Reading key (same as the interview PDF): in the preview, dark text is the fixed wording — say it as written; orange bold [CHANGE: …] / [fill: …] is the part you adapt for this professor, lab, company or question, with what to put there written inside the brackets.
Attached Documents
Upload any document for this session (PDF, Word, text, image — up to 10 MB each). Files stay in this tab's memory. Text files are read into this session's brief automatically; for a scanned PDF, paste its key points into the prepared material above.
AI Connections
Connect the AI you already have. Muse, ChatGPT, Gemini and Claude each connect with your own API key and become the brain for live answers — Muse uses a Meta Model API key from dev.meta.ai. Free options: the Gemini API key is free and answers live automatically; Muse, ChatGPT and Claude also work free through the copy-paste handoff on their cards — no key, no payment. Each workspace keeps its own keys separate. Switch any time, no restart.
Muse
Connect with your own Meta Model API key from dev.meta.ai ↗ (sign in → API keys → Create API key). Muse Spark then writes your live answers straight from this page. Your Muse app subscription does not cover this — API use is pay-as-you-go, billed by Meta separately from any plan.
No key yet? Get your Meta Model API key here ↗ — copy it, then paste it below.
Get an API keyFree option — no key, no payment: use the free Muse chat instead. One tap copies the question + your brief and opens Muse; paste it there, then paste the reply back in Live Copilot.
Open Muse ↗ChatGPT · OpenAI
Connect with an OpenAI API key from platform.openai.com ↗. Once pasted, live answers are generated by ChatGPT's models (gpt-6-luna for speed) straight from this page.
No key yet? Get your OpenAI API key here ↗ — copy it, then paste it below.
Get an API keyFree option — no key, no payment: use the free ChatGPT chat instead. One tap copies the question + your brief and opens the free chat; paste it there, then paste the reply back in Live Copilot.
Open free ChatGPT ↗Gemini · Google AI FREE KEY
Get your free key: 1) Open aistudio.google.com ↗ → 2) Get API key → 3) Create API key (free) → 4) Copy it and paste it below. Model used: gemini-3.5-flash-lite (Gemini Flash Lite — if that exact model is not on your key, the closest Flash Lite model is used automatically). The key stays in this browser tab's memory only. This is the best free live brain — it answers automatically, with no copy-paste.
No key yet? Get your free Gemini key here ↗ — copy it, then paste it below.
Get a free Gemini key ↗Not tested yet. “Test connection” really calls Gemini and shows “Gemini connected” or the exact error from Google.
Claude · Anthropic
Connect with an Anthropic API key from console.anthropic.com ↗. Once pasted, live answers are generated by Claude (Haiku for speed) straight from this page.
No key yet? Get your Anthropic API key here ↗ — copy it, then paste it below.
Get an API keyFree option — no key, no payment: use the free Claude chat instead. One tap copies the question + your brief and opens the free chat; paste it there, then paste the reply back in Live Copilot.
Open free Claude ↗Phone & QR — answers on your mobile
The safest place for answers is your phone, next to the laptop — nothing to capture on the shared screen at all. This build is prepared for copilot.appsnation.co: the QR code below is generated for that address, ready to scan once the page is uploaded there.
Works alongside your meeting app
"Works" means the copilot runs beside the meeting in a separate tab or on your phone, so nothing about the meeting app changes. Keep the copilot out of the shared tab and it is never part of what you share.
Privacy Shield
Completely yours. Here is exactly what this web version does for each privacy point — and, honestly, which ones only a native desktop app can fully do.
Invisible on Screen Share
Keep answers on your phone, or share a single tab/window — the copilot lives in a different tab, so it is never in the shared picture. Safe View swaps answers for plain meeting notes in one tap if you ever share the wrong thing.
In this web appInvisible in Dock
A web page has no Dock or taskbar icon of its own — it is just one browser tab among yours, with a plain title. Nothing extra appears in your Dock or taskbar.
In this web appInvisible in Activity Monitor
There is no separate process to find: the copilot runs inside your normal browser process, under the browser's own name. Only a native app can rename its process further.
Partly — browser process onlyInvisible to Tab Switching
Reading answers on your phone means you never switch tabs on the shared machine at all. On the laptop, Auto Safe View can hide answers the moment this tab loses focus.
In this web appCursor Undetectability
Nothing here changes your cursor, installs hooks, or injects overlays into other apps — your cursor behaves exactly as it always does, because the copilot never touches it.
In this web appFull-Screen Safety
If you present full-screen, present from the meeting app or a deck — keep this tab out of it and answers stay off-screen entirely. Entire-Screen sharing captures everything on that display, in any app, so the rule is simple: answers live on the phone.
In this web appBefore you share your screen — 20-second checklist
- Answers open on your phone, beside the keyboard — not on the laptop screen.
- Share a single tab or window, never Entire Screen.
- If you must use the laptop, turn on Safe View first (Ctrl/⌘ + Shift + H hides answers instantly).
- Close or mute notifications and any tab that shows the question feed.
- Do a dry run: share with a friend or a second device and confirm nothing of the copilot is visible.
Mock Test — Practice Interview
Add your attached documents or script, generate a full mock test, then practise in a voice-call demo session. Add as many questions as you want, or upload your own questions — answers and scoring follow your material.
1 · Your attached documents / script
Paste your attached documents or prepared script here, or upload it. Mock questions and suggested answers are built from this text only. This syncs with your Brief material.
Script key: dark text = say as written, from your material · orange bold [CHANGE: …] = adapt that part for this professor / lab / company / question.
Default is 20 questions. Change the number to generate as many as you need.
2 · Add your own questions
Type one question per line, or upload a questions file (.txt). Uploaded questions are added to the test and answered from your attached documents / script above.
Your mock test
No questions yet — generate 20 above or add your own.
Suggested answers use the interview-PDF style: dark text = say as written, from your material · orange bold [CHANGE: …] = the part to adapt for this professor / question.
Notes & Summary
Built from this session's transcript — what was asked, what you said, and what happens next.
Generate summary builds the summary, decisions and action items from this session's transcript turns only. The Word download is a Word-compatible document file made in your browser from that same summary — nothing is uploaded.
No turns yet. Answers you generate in Live Copilot will be summarised here.
Settings & Guide
Choosing the AI, the speech engine and the audio devices — plus the full guide: connecting your AI, your first meeting, hotkeys, screen sharing, speed and privacy. Items marked Desktop app (Electron) only describe the desktop build; this web page states its own limit under each one.
Settings — AI, speech & overlay
How the brain works: when an AI is connected (ChatGPT, Gemini or Claude), it answers live in natural human words — it reads your CV / documents and speaks like you would, never by pasting lines from them. With no AI connected, the Local fast engine answers instantly from your script / attached documents instead. Free chats (ChatGPT, Claude, Muse) work through copy-paste in AI Connections — no key needed — and Gemini's API key is free.
Settings → Audio devices
The same idea as in Zoom. Both start on Same as system, which follows your computer when you plug in headphones. Browsers hide the real device names until microphone access is allowed — press Allow mic & detect devices first: it asks for the mic, listens for a couple of seconds so you can see the live level, and then fills both lists with your actual microphones and speakers. The tests below then use exactly the mic and speaker you pick here.
Detection has not run yet on this visit — press Allow mic & detect devices. Until then the lists can only show “Same as system”, because the browser hides device names before mic permission. No device name is ever invented here: if a browser will not expose a list (for example Firefox does not list speakers), this line says so plainly.
No test run yet. Test speaker plays a short chime through the speaker picked above, then listens on the microphone picked above for 2 seconds and reports the level it actually measured — heard clearly, heard only faintly, or heard almost nothing. If the chime plays but the app hears nothing, the other person will not be transcribed, so this is the test that matters. (In this web page the check is a microphone hearing check, not a true speaker loopback — and a headset's own mic usually cannot hear its own speakers, so with a headset this test can read “not heard” even when your voice is fine.) Test mic records you for 4 seconds on the microphone picked above and tells you if you came through clearly.
What the picks control: the level meters and these tests use exactly the microphone selected above, and the chime is sent to the selected speaker where the browser allows output selection (Chrome). One honest limit: the live speech recogniser (Live Copilot / Mock Test) always uses the browser's own microphone — normally your system default — so set this same mic as your computer's default input for calls. “Same as system” stays the safe default whenever a list is unavailable.
Connect your AI — it looks for Claude and ChatGPT by itself
When the desktop app opens it looks for Claude and ChatGPT on your computer by itself:
| Situation | What happens |
|---|---|
| Already signed in to Claude or ChatGPT | It connects straight away. Nothing to do. |
| Signed in only to the one you did not pick | It switches to that one by itself. |
| Signed in to neither | Press Sign in with Claude or Sign in with ChatGPT, log in in the browser window that opens, and it connects as soon as you are done. |
Web page limit: this page cannot detect or reuse your Claude / ChatGPT sign-in — that auto-detect is Desktop app (Electron) only. Here, use AI Connections: ChatGPT, Gemini and Claude connect with your own API key, Muse through the chat handoff.
The first time on a Mac — macOS asks for two permissions. Allow both for Electron: Microphone, to hear you; Screen & System Audio Recording, to hear the other person (no video is recorded). If macOS offers Quit & Reopen after you allow it, quit, then open Meeting Copilot again. Desktop app (Electron) only — a web page instead asks for microphone permission in the browser.
Your first meeting — before, during, after
- Before the call: write the brief. In the dashboard, open Brief and fill in who you are meeting, your role and what a good outcome looks like. Paste anything the answers may draw on: your pitch, your prices, past results, your attached documents — or press Generate script from my attached documents there and the script is built in the SAY / NOTE - DON'T SAY format for you. Then press Ask the AI what else it needs and answer its questions. That minute improves the live answers more than anything else. Everything you type also helps the speech recognition: words from your brief, like product names and jargon, come out spelled correctly.
- During the call: press Start listening. The overlay shows: what they are saying, as it is transcribed; SAY THIS, the suggested answer, appearing as it is written; three questions you could ask next; and a box to ask the AI anything directly (the question box in Live Copilot).
- After the call: open Notes & summary. Press Generate summary for a summary, decisions, action items and a follow-up message, then Download Word (.docx).
Hotkeys & the overlay
| Keys | What it does |
|---|---|
| Ctrl Shift Space | Show or hide the overlay |
| Ctrl Shift H | Hide it instantly |
| Ctrl Shift X | Click-through: the mouse passes through the overlay |
| Ctrl Shift A | Jump to the ask box |
| Ctrl Shift R | Write that answer again |
On a Mac these use the Control key. These hotkeys work while this page is focused — registering them across the whole system is Desktop app (Electron) only. Drag the overlay (the answer popup) from its head bar. S / M / L resize it and the slider in Settings above sets how see-through it is.
Screen sharing — what the other side sees
Share a window or a tab, not your entire screen. Then the overlay is not part of what the other side sees, on Windows and on Mac. (Desktop app behaviour:)
| What you share | Mac | Windows |
|---|---|---|
| A window or a browser tab | Hidden | Hidden |
| Your entire screen | Hidden on recent macOS | Hidden on most PCs; on some older graphics chips it shows as a dark box, never as readable text |
When you share your entire screen from Chrome on Windows (Google Meet, or Zoom and Teams in the browser), the desktop app notices and hides the overlay by itself — try exactly that in the Meeting simulation in Live Copilot. The desktop Zoom and Teams apps do not announce a share the same way, so there, share a window.
Web page limit, stated plainly: this web page is not the desktop app — if the popup is on the same display and you share the entire screen, it will be captured unless the simulation's auto-hide (or Safe View) has hidden it first. For complete peace of mind in any share mode, use the phone view: Phone & QR → scan the QR code with a phone on the same Wi-Fi.
Speed — how fast answers arrive
Two things take time after the other person stops talking: turning their speech into text, and writing the answer.
| Speech to text | Words on screen |
|---|---|
| Cloud (Deepgram, free tier available) | About 2 s |
| On your computer, modern laptop | About 3 to 4 s |
| On your computer, 2012 laptop | About 5 s |
Speech to text runs on your computer by default, with no account and no cost. For the fastest answers, get a free key from deepgram.com and set Settings → Speech backend → cloud. If the connection drops mid-call it falls back to your computer by itself. On a newer machine, Settings → Local model → base.en is more accurate.
Privacy
- Your account, brief, transcripts and settings are saved only in this browser on this device (browser storage — the web equivalent of the desktop app's own data/ folder) and are never uploaded by this page. API keys are the exception: they stay in this tab's memory only and are never written to storage. Nothing syncs to another device or browser.
- Speech to text runs on your computer unless you choose the cloud option (in this web page, the browser's own recogniser is used).
- The only thing that leaves your computer is what the AI needs to write an answer: the brief and the recent conversation, sent to Claude or ChatGPT (or Gemini) under your own account / API key, only when you connected one.
- The phone view is reached through your own hosting link and its QR code — use it on your own Wi-Fi / your own devices.
- Use it responsibly. In many places you must tell people before you transcribe a call, and some workplaces, schools and exams do not allow assistants like this. Check the rules that apply to you.
Tutorials
Short walkthroughs for every part of Ghost Copilot. Each one takes under a minute to follow.
1 · Start your first session
Go to Call Sessions → Create Session → open Brief & Documents and check your prepared material → press Start Session. The session turns Live and the copilot is ready.
2 · Get a fast answer
In Live Copilot, type what they just asked (or press Start listening). The first 1–2 lines appear first — start speaking from them while the full SAY THIS script streams in behind. If the question changes, the answer switches with it.
3 · Use the answer popup
While a session is live, the answer popup opens on its own the moment a new answer is ready, with the latest answer in large text. You can also open it as a separate popup window from the Live Copilot header, and drag that window to a second screen.
4 · Read answers on your phone
Open Phone & QR, generate the QR code for copilot.appsnation.co, and scan it before the call. On the phone, Phone answer view shows only the big SAY THIS script — the safest place for answers.
5 · Safe View in one keystroke
Press Ctrl/⌘ + Shift + H at any moment and every answer — including the popup — is replaced by a plain meeting-notes page. Press it again to bring answers back.
6 · After the call
End the session from Live Copilot, then open Notes & Summary for the full transcript, what was covered, next steps, and a copy-ready follow-up message.
7 · Mock Test with voice practice
Open Mock Test, paste or upload your attached documents / script, choose how many questions you want (20 by default) and generate them. Add your own questions or upload a questions file, then Start demo voice session — each question is spoken aloud, you answer by voice or typing, and a full scored result appears at the end.
Support
Help with Ghost Copilot by AppsNation. Start with the quick answers below — most issues are one of these three.
The answer popup did not open
The popup only appears while a session is Live and "Auto-popup" is switched on in Live controls. If the separate popup window is blocked, allow popups for copilot.appsnation.co once in your browser — the floating answer card still appears on its own either way.
The mic is not hearing the question
Speech recognition needs Chrome and microphone permission, and the page must be open over https. The mic hears this device only — play the call on speaker, or type the question; answers are just as fast typed.
My AI key stopped working
Keys live only in this tab's memory, so closing the tab erases them — reconnect under AI Connections. If a connected AI fails mid-call, the local fast engine answers automatically so you are never left without a script.
Still stuck?
Tap "Copy a support message", paste it into your email or chat, add what you were doing when the problem happened, and send it to the AppsNation team. Include your browser name and whether you were on the laptop or the phone view.