[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-93405":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":9,"language":10,"languages":9,"totalLinesOfCode":9,"stars":11,"forks":12,"watchers":13,"openIssues":14,"contributorsCount":15,"subscribersCount":15,"size":15,"stars1d":16,"stars7d":17,"stars30d":17,"stars90d":15,"forks30d":15,"starsTrendScore":18,"compositeScore":19,"rankGlobal":9,"rankLanguage":9,"license":20,"archived":21,"fork":21,"defaultBranch":22,"hasWiki":23,"hasPages":21,"topics":24,"createdAt":9,"pushedAt":9,"updatedAt":25,"readmeContent":26,"aiSummary":27,"trendingCount":15,"starSnapshotCount":15,"syncStatus":28,"lastSyncTime":29,"discoverSource":30},93405,"cue","Blueturboguy07\u002Fcue","Blueturboguy07","Open-source macOS AI copilot that floats over your screen, sees\u002Fhears your meetings, and stays hidden from screen shares. Cluely alternative, bring-your-own-key.",null,"JavaScript",679,144,113,1,0,33,566,214,100.48,"GNU General Public License v3.0",false,"main",true,[],"2026-07-22 04:02:09","\u003Cdiv align=\"center\">\n\n# cue\n\n**An open-source AI copilot that floats over your screen — sees what you see, hears your meetings, and stays hidden from screen shares.**\n\nA free, self-hosted alternative to Cluely. Bring your own AI key (OpenAI · Anthropic · Google Gemini).\n\n\u003Cimg src=\"docs\u002Ftutorial.png\" width=\"620\" alt=\"cue first-run tutorial\" \u002F>\n\n\u003C\u002Fdiv>\n\n---\n\n> [!IMPORTANT]\n> **Please read this first.** cue tries to stay out of screen recordings\u002Fshares, but this is **best-effort, not guaranteed** — on macOS 15.4+ Apple can let modern capture tools see it anyway, and a phone camera always can. Using a hidden assistant during a **proctored exam, job interview, or recorded meeting** may break that platform's rules and, in some places, consent laws. cue is built for legitimate uses — your own notes, studying, accessibility, and practice. **You are responsible for how you use it.**\n>\n> On Zoom specifically, whether cue is hidden depends on one setting — **Settings → Share Screen → Screen capture mode → \"Advanced capture with window filtering.\"**\n>\n> \u003Cimg src=\"docs\u002Fzoom-capture-mode.png\" width=\"560\" alt=\"Zoom Settings → Share Screen → Screen capture mode set to Advanced capture with window filtering\" \u002F>\n\n---\n\n## What it does\n\ncue floats a small glass panel on top of everything. It takes **three separate inputs** — your **screen**, your **microphone**, and your **meeting audio** (what the other person says) — and uses an AI model to help you in real time.\n\n| Feature | How to trigger | What it uses |\n|---|---|---|\n| **Assist** | `⌘` `↵` or the *Assist* button | your screen + recent conversation |\n| **What should I say?** | button | meeting audio + your mic |\n| **Follow-up questions** | button | the whole conversation |\n| **Recap** | button | the whole conversation |\n| **Ask anything** | type + `↵` | your screen + conversation |\n| **Solve a coding problem** | `⌘` `H` | your screen only |\n| **Smart** toggle | pill in the box | switches to a smarter (slower) model |\n\nIt's a copilot for **live meetings** (\"what do I say to that?\") and **coding problems** (screenshot → full solution), and it's designed to be **invisible in screen shares** so it stays your private assistant.\n\n---\n\n## Install\n\nThere are two ways to install cue. **If you're not a developer, use Option A.**\n\n### Option A — Download the app (easiest)\n\n1. Go to the [**Releases**](..\u002F..\u002Freleases) page and download **`cue-mac.zip`**.\n2. Double-click the zip to unzip it. You'll get **`cue.app`**.\n3. Drag **`cue.app`** into your **Applications** folder.\n4. **First open (important):** because cue is a free app without a paid Apple certificate, macOS will refuse to open it normally the first time. Do this once:\n   - **Right-click** `cue.app` → **Open** → click **Open** in the dialog.\n   - If macOS instead says **\"cue is damaged and can't be opened,\"** open the **Terminal** app and paste this line, then press Return:\n     ```bash\n     xattr -cr \u002FApplications\u002Fcue.app\n     ```\n     Then double-click cue.app again. (This just tells macOS \"yes, I trust this app I downloaded.\" It's safe.)\n\nAfter that, cue opens normally forever.\n\n### Option B — Run from source (developers)\n\nYou need [Node.js](https:\u002F\u002Fnodejs.org) 18+ installed. No Xcode required.\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002FBlueturboguy07\u002Fcue.git\ncd cue\nnpm install\nnpm start\n```\n\nTo build your own `cue.app`:\n```bash\nnpm run pack      # creates dist\u002Fmac-arm64\u002Fcue.app\n```\n> Note: the packaged app is **ad-hoc signed** (no paid Apple certificate). macOS ties permission grants to the exact build, so **rebuilding resets the mic\u002Fscreen permissions** — you'll grant them again. For everyday use, build once and keep it.\n\n---\n\n## First launch — the 1-minute setup\n\nWhen cue opens the first time, a **built-in tutorial** walks you through everything below. You can reopen it anytime by clicking the **cue logo** (top-left of the pill). Here's the same thing in writing.\n\n### Step 1 — Grant two macOS permissions\n\ncue can't help until macOS lets it see and hear. When you first use a feature, macOS will prompt you — click **Allow**. If a prompt doesn't appear, add cue manually:\n\n- **Microphone:** System Settings → **Privacy & Security** → **Microphone** → turn on **cue**.\n- **Screen Recording:** System Settings → **Privacy & Security** → **Screen Recording** → turn on **cue**. (This one grant covers both screenshots *and* meeting audio.) macOS may ask you to **quit & reopen** cue — let it.\n\n### Step 2 — Add your AI key (bring your own)\n\ncue uses **your own** API key, so it's free to run (you only pay your AI provider for what you use). Click the **`...`** button in the input box (or press `⌘` `,`) to open **Settings**, pick a provider, and paste your key:\n\n| Provider | Get a key | Notes |\n|---|---|---|\n| **OpenAI** | [platform.openai.com\u002Fapi-keys](https:\u002F\u002Fplatform.openai.com\u002Fapi-keys) | One key does everything — **but** for the *listening* features the key must have **Whisper \u002F audio** access (a \"restricted\" project key that only allows chat will give a 403 on transcription). |\n| **Anthropic (Claude)** | [console.anthropic.com](https:\u002F\u002Fconsole.anthropic.com) | Great for screen & coding help. Claude has no speech-to-text, so add an OpenAI or Gemini key too if you want the listening features. |\n| **Google Gemini** | [aistudio.google.com\u002Fapikey](https:\u002F\u002Faistudio.google.com\u002Fapikey) | One key does chat + transcription. |\n\nYour key is stored **only on your computer** (in `cue-data.json`) and is sent **only** to that provider. cue has no server and collects nothing.\n\n### Step 3 — The Zoom setting (only needed for Zoom)\n\ncue is hidden from most screen-share tools automatically — **Google Meet, Microsoft Teams, and QuickTime need nothing.** **Zoom** has a specific setting that decides whether it respects cue's \"don't capture me\" flag:\n\n> **Zoom → Settings → Share Screen → Advanced → Screen capture mode → choose \"Advanced capture with window filtering.\"**\n\n\u003Cdiv align=\"center\">\u003Cimg src=\"docs\u002Fzoom-setting.png\" width=\"560\" alt=\"Zoom screen capture mode setting\" \u002F>\u003C\u002Fdiv>\n\n**Why:** the *\"...with window filtering\"* modes tell Zoom to leave out windows that mark themselves as private — which is exactly what cue does. The **\"Advanced capture without window filtering\"** mode grabs the raw screen and **will show cue**, so avoid it.\n\n---\n\n## How to use it\n\n- **`⌘` `↵` — Assist.** The do-the-smart-thing key. On a coding problem it solves it; in a conversation it tells you what to say. Works from anywhere.\n- **`⌘` `H` — Solve what's on screen.** Screenshots a coding problem and returns the approach, code, and time\u002Fspace complexity.\n- **The `▢` button** (top bar) — start\u002Fstop **listening** to a meeting. The green dot means it's live.\n- **Type a question** in the box and press `↵` to ask about your screen or conversation.\n- **Smart** — flip it on for a smarter, more thorough model; off for fast and cheap.\n- **Hide** collapses the panel to just the top bar. Drag cue around by the **top pill**. Quit with `⌘` `⇧` `X`.\n\nThe panel is see-through and click-through — the empty space around it never blocks the app behind it.\n\n---\n\n## How it works (under the hood)\n\ncue is an [Electron](https:\u002F\u002Fwww.electronjs.org\u002F) app. Everything runs locally except the calls to your chosen AI provider.\n\n**The three inputs are kept completely separate:**\n- **Screen** — captured with Electron's `desktopCapturer` (full-resolution screenshots, taken only when a feature needs one).\n- **Your mic (\"You\")** — `getUserMedia` → downsampled to 16 kHz audio → transcribed.\n- **Meeting audio (\"Them\")** — `getDisplayMedia` loopback capture of your system's output audio, kept on its own channel so cue knows *who* said what.\n\nBoth audio streams are transcribed (OpenAI Whisper or Gemini) and fed, with an optional screenshot, to your AI model. Responses **stream** into the panel word-by-word.\n\n**The invisibility** is a single macOS window flag: `setContentProtection(true)`, which sets `NSWindowSharingNone`. This asks the window server to exclude cue from screen-capture streams. It's the same mechanism DRM apps and Zoom's own toolbar use. It is **not** a GPU trick or a special overlay layer — and on macOS 15.4+ Apple lets some capture tools ignore it, which is why it's best-effort (see the disclaimer at the top).\n\n```\nmain process ──┬─ overlay window (frameless, transparent, always-on-top, content-protected)\n               ├─ screenshot capture (desktopCapturer)\n               ├─ speech-to-text (Whisper \u002F Gemini)      ── \"You\" + \"Them\" channels\n               └─ LLM streaming (OpenAI \u002F Anthropic \u002F Gemini)\nrenderer ──────┴─ the glass UI + mic capture + system-audio loopback\n```\n\n---\n\n## Troubleshooting\n\n**\"It says give access, but I already gave access.\"**\nYou probably granted an older build. Because the app is ad-hoc signed, a rebuild changes its identity and macOS stops honoring the old grant (the checkmark can linger). Toggle cue **off and on** in System Settings → Screen Recording, or remove and re-add it.\n\n**A feature returns \"403\" \u002F \"no access to model.\"**\nYour API key is restricted. Most often it's an OpenAI **project key that only allows chat models** — it works for screen\u002Fcoding help but 403s on transcription (Whisper). Fix: enable audio\u002FWhisper on the key, use an unrestricted key, or add a Gemini key (cue falls back to it for transcription).\n\n**Listening does nothing \u002F no transcript.**\nCheck Settings shows a transcription-capable key (OpenAI with Whisper, or Gemini). Also make sure Screen Recording is granted (meeting audio needs it).\n\n**cue shows up in my Zoom share.**\nSet Zoom's **Screen capture mode** to *\"Advanced capture with window filtering\"* (see Step 3). And remember: on macOS 15.4+ this can still fail — it's best-effort.\n\n**\"cue is damaged and can't be opened.\"**\nRun `xattr -cr \u002FApplications\u002Fcue.app` in Terminal once (see Install → Option A).\n\n---\n\n## Privacy\n\n- No accounts, no servers, no telemetry. cue collects nothing.\n- Your API keys live in a local file (`cue-data.json`) and are sent only to the provider you chose.\n- Screenshots and audio are sent to your AI provider only when a feature runs, and are not stored by cue beyond the current session's transcript (kept in memory).\n\n## Contributing\n\nIssues and PRs welcome. cue is intentionally small and readable — `main.js` (app + capture + AI), `renderer\u002F` (the UI), `src\u002F` (providers). No build step for the source (plain HTML\u002FCSS\u002FJS).\n\n## Credits & license\n\nBuilt as an open-source study of how tools like **Cluely** and **Interview Coder** work. Modeled on the open-source clones `pickle-com\u002Fglass` and `sohzm\u002Fcheating-daddy`.\n\n**License: [GPL-3.0-or-later](LICENSE).**\n","cue 是一款开源的 macOS AI 辅助工具，以浮动玻璃面板形式实时感知屏幕内容、本地麦克风及会议音频输入，并在屏幕共享中尽力保持隐藏。它支持自选大模型（OpenAI\u002FAnthropic\u002FGemini），提供会议实时应答建议、对话复盘、编码问题求解等 Copilot 功能，所有处理可本地完成。项目采用 JavaScript 开发，强调隐私与可控性，适用于个人学习笔记、无障碍辅助、会议准备及编程调试等非受控、非合谋场景。",2,"2026-07-17 02:30:11","CREATED_QUERY"]