AceTheScreen — User Manual

A Windows desktop overlay for interview preparation: dictate or type questions, see live transcription, and get AI replies on top of your screen while you practice or participate in calls.

Table of contents

1. Overview
2. System requirements
3. Installation
4. Quick start (5 minutes)
5. Main interface
6. Keyboard shortcuts
7. Dictation and transcript
8. AI providers
9. Settings reference
10. Subscription and billing
11. Screen sharing and privacy
12. Built-in local AI setup
13. Screenshot vision (Ollama)
14. Window behavior
15. Troubleshooting
16. FAQ

1. Overview

AceTheScreen runs as a transparent, always-on-top window on Windows. It:

• Converts your speech to text using Windows built-in dictation
• Sends only new text (typed or dictated since your last AI reply) to an AI provider
• Shows the full transcript and AI reply in a compact overlay
• Keeps settings behind a separate panel so you can hide the overlay during screen shares

Plans at a glance

2. System requirements

• OS: Windows 10 or 11 (64-bit)
• Microphone: Required for dictation (built-in or USB headset)
• Internet: Required for OpenAI; optional for local AI after setup
• OpenAI: Your own API key from platform.openai.com
• Local AI (paid): ~0.5–1 GB disk space for a small model; more RAM = faster replies
• Ollama (paid, optional): Ollama installed separately for legacy mode and screenshot vision

3. Installation

1. Download and extract the Windows build (or run the built .exe from your release folder).
2. Launch the application. The window is transparent and may not appear on the taskbar by default.
3. Press F2 to open Settings if you do not see the overlay.
4. Press S to open the main panel (transcript + AI reply).

Tip: If you cannot find the window, use Alt+Tab or press F4 (default) to toggle “always on top” so it stays above other apps.

4. Quick start (5 minutes)

1. Open Settings — press F2.
2. Choose OpenAI (cloud) — this is the default and works on the free plan.
3. Paste your OpenAI API key and confirm the model (default: gpt-4o-mini).
4. Optionally fill in Discussion context (warmup) — e.g. “Senior backend interview, focus on C# and system design.”
5. Click Save settings.
6. Press S to open the main panel.
7. Press F10 to start recording (dictation). Speak your question.
8. Press F10 again to stop. New text is sent to the AI automatically if Auto-send when recording stops is enabled.
9. Read the AI reply in the main panel.

5. Main interface

Mini HUD (bottom-left)

When the main panel is closed, a compact overlay shows recent session chat messages (your sends and AI replies), plus errors or recording status when active.

Main panel (press S)

Settings panel (press F2)

Full configuration: hotkeys, AI provider, billing, system prompt, warmup context, and opacity. Opens from F2 only (not the main S panel).

6. Keyboard shortcuts

Default keys (all rebindable in Settings):

Rebinding hotkeys

1. Open Settings (F2).
2. Click the button next to the action you want to change.
3. Press the new key. Press Esc to cancel.
4. Click Save settings.

Allowed keys include F1–F12, letter keys, and a few symbols (see the rebind UI).

Window controls (separate from overlay hotkeys)

These are configured on the transparent window component in the Unity project/build defaults:

7. Dictation and transcript

How recording starts

Listening can begin in two ways:

1. Automatically — when Auto-start listening after AI reply is enabled (default: on), dictation starts again as soon as the AI finishes a reply (warmup-only requests are excluded).
2. Manually — press your Record hotkey (F10 by default) to start or stop dictation at any time.

If you are already recording when the AI reply arrives (e.g. with stream-while-recording), the app does not restart the mic — it keeps listening.

Press the record hotkey again to stop dictation; with Auto-send when recording stops enabled, new speech is sent to the AI, then listening can auto-start again after the reply.

How recording works

1. Press your Record hotkey (F10 by default), or wait for auto-start after an AI reply.
2. Windows dictation listens through your default microphone.
3. Live text appears in the dictation log (committed phrases plus in-progress “hypothesis” text).
4. Press Record again to stop. The app finalizes any pending speech and updates the dictation log.

Incremental AI sends (dictation)

The dictation log keeps all speech, but dictation-driven AI requests send only new content since the last successful reply:

• If you dictate “What is polymorphism?” and get a reply, then dictate “Give an example in C#”, only “Give an example in C#” is sent as the latest user message.
• Each successful dictation send also appears as a You message in session history.
• Compose sends (Send button / Ctrl+Enter) send exactly what you typed — they do not use the unsent-dictation logic.

Session chat history in AI requests (Settings)

When Include session chat history in AI requests is enabled (default: on), the app also sends up to N prior turn pairs from the current session (your message + AI reply) with each request. That lets follow-ups like “expand on that” work across OpenAI, built-in local, and Ollama.

• Default: 6 prior pairs (12 messages) plus the current message.
• Disable this to reduce tokens/cost or if you want each request treated in isolation.
• Click Save settings after changing the turn count.

Recording options (Settings)

Manual send

Press F11 (or Send dictation to AI in Settings) to send unsent dictation text at any time. Use the compose box in the main panel to type and send separate messages (Enter or F9).

Copy and export session chat

• Click any message in Session history to copy it to the clipboard.
• Click Copy session to copy the entire current session (all messages).
• Click Clear session to start a fresh chat log (dictation log is kept).

Collapse dictation log

Uncheck Show dictation log on the main panel to hide the speech log and expand session history. Your preference is saved automatically.

Requirements for dictation

• A working microphone set as the default recording device in Windows Sound settings.
• Windows only — dictation is not available on macOS or Linux builds.

Presentation Q&A mode

Turn on Presentation Q&A mode in Settings to handle questions from an audience.

1. Describe your talk in Discussion context (topic, key facts, numbers, terms) and click Warm up AI. The AI treats this as the source of truth.
2. Press the record hotkey. The app listens continuously.
3. When the speaker pauses, the app checks what was just said. If it sounds like a question, it asks the AI and shows Q: … followed by a short answer you can read out.
4. If a question was missed, press F8 (Answer last-heard question) to answer whatever was just heard.

Tips and limits:

• Dictation hears your default microphone. In a room this picks up the audience. For remote attendees, route meeting audio to the dictation input (Stereo Mix or a virtual audio cable).
• The app can't tell speakers apart, so a question you ask yourself may trigger an answer.
• The AI is told not to invent facts. If it isn't sure, it suggests saying you will follow up. Always check answers before relying on them.

Answer styles

Choose how the AI answers: Clean-up (tidy dictation), Concise, STAR (behavioural interview structure), Technical (definition, key points, trade-offs) or Coding (approach, code, complexity). A custom system prompt, if you set one, is kept and the style is added to it.

Saving a session

Use Save .txt above the session history to write the conversation to Documents\InterviewOverlay.

8. AI providers

Select a provider in Settings → provider buttons at the top of the AI section.

OpenAI (cloud) — free / default

• Requires: OpenAI API key (sk-…)
• Model: Configurable (default gpt-4o-mini)
• Best for: Fast setup, strong answers, no local disk use
• Cost: Billed by OpenAI to your API account

Built-in local (fast) — paid

• Requires: Active subscription; llama-server.exe + a .gguf model (see Section 12)
• Best for: Low latency, offline use, no per-token cloud cost
• Note: Text only — no screenshot vision

Ollama (legacy) — paid

• Requires: Active subscription; Ollama running locally (default URL http://localhost:11434)
• Default text model: phi3:mini
• Best for: Screenshot vision and users who already use Ollama
• Note: Often slower than built-in local AI on first load

System prompt

Default behavior: clean up raw speech-to-text — light punctuation, fix obvious mishearings, do not invent content. You can replace this in Settings with your own instructions (e.g. “Answer as a senior engineer in a technical interview”).

Discussion context (warmup)

Describe the interview, role, stack, or topic. This text is prepended to every AI request so replies stay on context.

Warm up AI sends a short confirmation request so the model acknowledges the context before you start dictating.

9. Settings reference

Hotkeys

Rebind main panel, settings, record, send to AI, and screenshot keys.

Hide overlay while sharing

When enabled, the transcript/reply HUD and main panel are hidden. Settings remain available via F2. Use this before Google Meet, Zoom, or other screen shares so the overlay is not visible on your shared screen.

This is intentional hiding — it does not make the window invisible to the OS or all capture methods. It simply stops drawing the overlay UI.

AI provider block

Provider-specific fields appear below the provider selector (API key, local engine status, Ollama URL/model, max reply tokens, vision model).

Billing / Subscription

See Section 10.

System prompt & Save

• Save settings — persists hotkeys, AI config, billing fields, opacity, and dictation options.
• Send text to AI — manual send from Settings.
• Screenshot → Ollama — same as F12 hotkey.

Opacity

• Panel opacity — background of the overlay boxes.
• Text opacity — transcript and reply text only (useful for a subtle HUD).

10. Subscription and billing

OpenAI is free to use in the app (you still pay OpenAI directly for API usage). Built-in local, Ollama, and screenshot vision require an active subscription verified through your seller’s website.

Get your license key

1. Subscribe at acethescreen.com.
2. Your license key (looks like ATS-XXXX-XXXX-XXXX-XXXX) is emailed right after payment. You can also see it any time in your account — we email you a one-time sign-in link, no password needed.

Activate in the app

1. Open Settings (F2) → Billing / Subscription.
2. Paste your license key.
3. Click Verify. Plan status should show Subscribed.

The app re-checks your license automatically and keeps working offline for up to 7 days between checks. One license works on up to 2 devices; remove old devices from your account page.

Manage or cancel

Use Manage billing & cancel in your account to update your card, download invoices or cancel. Access continues until the end of the paid period.

After subscribing

• You can select Built-in local or Ollama in Settings.
• Without an active subscription, the app reverts to OpenAI if a paid provider was selected.

11. Screen sharing and privacy

Hide overlay while sharing

1. Open Settings (F2) before sharing your screen.
2. Enable Hide transcript/reply overlay while sharing screen.
3. The mini HUD and main panel disappear; open Settings again with F2 to turn sharing mode off.

Hide from screen capture

Enable Hide this window from screen capture / recording in Settings. The overlay stays visible on your monitor but is left out of screen shares, recordings and screenshots made by most software (requires Windows 10 version 2004 or later). A phone camera pointed at the screen can still see it, and some capture tools ignore the setting. Only use this where it is permitted — see the Terms of Service.

What this does and does not do

• Does: Hides the overlay UI so it is not shown in your normal workflow during a share.
• Does not: Guarantee invisibility to every screen-capture or streaming tool. Treat sensitive content accordingly.

Taskbar icon

The app can run hidden from the taskbar. Use Alt+Tab or your hotkeys to bring it back.

12. Built-in local AI setup

Paid subscribers can use Built-in local for fast, offline text replies via llama-server and a GGUF model.

Option A — Bundled with the app (recommended)

Files ship under StreamingAssets/LocalAi/:

LocalAi/
  llama-server.exe          ← from llama.cpp Windows release
  Models/
    your-model.gguf         ← small instruct model (e.g. Qwen 0.5B)

Recommended model: qwen2.5-0.5b-instruct-q4_k_m.gguf (~0.4 GB)

Downloads:

• Engine: llama.cpp releases (Windows x64 zip)
• Model: Qwen2.5 0.5B Instruct GGUF

If you received a developer build, the seller may have already bundled these files.

Option B — Runtime download (if enabled)

1. Open Settings → select Built-in local.
2. Click Download engine, then download a model from the list (e.g. Qwen 0.5B).
3. Click Use on the installed model.

If runtime download is disabled, you must add files manually under StreamingAssets/LocalAi/ (see SETUP.txt in that folder).

GPU acceleration

In the Unity LocalAiModelSetup asset (developer/seller config), GPU layers controls offloading to GPU:

• 0 = CPU only
• Higher values = more layers on GPU (requires a GPU build of llama-server, e.g. Vulkan/CUDA)

Verify local AI is ready

In Settings, the built-in local section shows:

• engine OK / engine missing
• model OK / model missing

Fix any missing items before sending to AI.

13. Screenshot vision (Ollama)

Capture what is on screen and ask Ollama’s vision model to describe it (useful for coding questions, diagrams, or slides).

Requirements

• Paid subscription
• Provider set to Ollama (legacy) — not Built-in local (text-only)
• Ollama running with a vision model (default: moondream)

ollama pull moondream

Usage

1. Configure Ollama URL, vision model, and optionally Include transcript text when sending a screenshot.
2. Press F12 or click Screenshot → Ollama in the main panel or Settings.
3. Wait for the vision reply (first run can be slow while the model loads).

The screenshot captures the game view (your overlay window content). Position relevant material on screen before capturing.

14. Window behavior

Transparency

The overlay uses a transparent Windows window. Areas without UI let you see and click through to apps behind (depending on click-through mode).

Click-through

• F1 toggles whether the window receives mouse clicks.
• In automatic mode, clicks go to the overlay only when the cursor is over panel UI; otherwise they pass to the desktop.
• Left Alt (hold) temporarily passes the mouse through.

Always on top

F4 toggles whether the overlay stays above normal windows.

Mouse interaction while editing

When the main panel or Settings is open, the overlay accepts mouse input for sliders, text fields, and buttons.

15. Troubleshooting

“No recording device found”

• Connect a microphone and set it as the default input in Windows Settings → System → Sound.
• Restart the app after changing devices.

Dictation does nothing / no speech detected

• Confirm the mic is not muted in Windows or on the device.
• Speak clearly; check that another app is not exclusively using the mic.
• Open the main panel (S) and watch the status line while recording.

“API key is empty”

• Open Settings, paste your OpenAI key, click Save settings, or switch to Built-in local (paid).

“Paid plan required…”

• Complete checkout and click Verify in Billing settings, or contact support if you already paid.

Local AI: “engine missing” / “model missing”

• Copy llama-server.exe and a .gguf file into StreamingAssets/LocalAi/ as described in Section 12, or use Download engine / model buttons if runtime download is enabled.

OpenAI errors / slow responses

• Check your API key and billing at OpenAI.
• Try a smaller/faster model (e.g. gpt-4o-mini).
• Check your internet connection.

Ollama errors

• Ensure Ollama is running: ollama serve or the Ollama tray app.
• Pull the model: ollama pull phi3:mini (text) or ollama pull moondream (vision).
• Confirm the URL in Settings matches your Ollama host (default http://localhost:11434).

AI sends duplicate or old text

• The app tracks text sent at the last successful reply. If you edit the transcript heavily, only the unsent suffix is sent. To resend everything, clear the transcript or edit so the unsent portion includes what you want.

Cannot find the app window

• Press Alt+Tab.
• Press F2 for Settings or S for the main panel.
• Press F4 to ensure the window is topmost.

Overlay still visible when sharing

• Enable Hide overlay while sharing in Settings before starting the share.

16. FAQ

Do I need an OpenAI account? For the free/default path, yes — you need an API key. Local providers (paid) do not use OpenAI.

Is my API key stored securely? Keys are saved locally in Unity PlayerPrefs on your PC. Do not share your build folder or config with others.

Can I use this on Mac or Linux? The overlay and Windows dictation target Windows standalone builds. Other platforms are not supported for dictation.

Does the transcript persist after closing the app? The last AI reply is saved. Transcript persistence depends on your build; re-open the main panel to review or continue typing.

Can I use ChatGPT without dictation? Yes. Open the main panel (S), type in the Message to AI compose box, and press Enter, click Send, or press F9.

Why only send new text to the AI? So follow-up questions feel like a natural conversation without resending the entire history each time, while you still see the full transcript locally.

Which local option should I use? Prefer Built-in local for speed. Use Ollama if you need screenshot vision or already run Ollama models.

Support

For subscription, license and activation issues, email support@acethescreen.com from your purchase email address and include your license key.

For OpenAI billing and API limits, see OpenAI Help.

Document version: matches the AceTheScreen Windows app, including Presentation Q&A mode, answer styles, streaming replies, capture exclusion and license-key activation.