Watch what you browse and chat about. Predict the next tasks. Act on them.
Cue turns Chrome into an intent-aware collaborator. With Go Live on, it captures browsing trajectory, searches, and AI chats (ChatGPT, Claude, Gemini, Perplexity), infers what you’re converging toward, and proactively suggests the next executable tasks—Gmail drafts, calendar events, docs, and build prompts for Antigravity or Cursor—through the on-page Halo strip and the Cue dashboard.
It is not primarily a meeting recorder. Optional session recording still exists for mic capture and transcription, but the core loop is context → intent → predicted tasks → action.
- Go Live context tracking — Tracks URL trajectory, dwell time, searches, and page category changes while you work
- AI chat capture — Pulls conversation context from major AI chat sites into a shared context store
- Intent & idea convergence — Scores whether you’re converging on a topic or build intent, and biases suggestions accordingly
- Predicted task suggestions — Gemini-backed
/suggest_tasksturns live context into structured, MCP-ready next actions (accept / dismiss) - Halo Strip UI — Always-available toolbar: Ask AI, Go Live, suggested tasks,
@gmail/@calendar/@docsstyle commands, voice wake settings - Executable Workspace actions — Preview and run Gmail, Calendar, Tasks, Docs, Drive, and Sheets actions via MCP
- Antigravity / Cursor prompts — When build intent is high, generate a structured, paste-ready prompt from your research context
- Dashboard — Activity feed for suggested tasks, Mosaic for browsing/intent insights, plus a session library
- Optional session recording — Mic capture → transcription / summary when you want a recorded session (secondary to Go Live)
- Go Live — Start tracking from the Halo strip
- Browse and chat — Cue builds context from sites, searches, and AI conversations
- Get predictions — Auto-suggestions appear when intent shifts or converges
- Accept or execute — Approve tasks or run
@commands into Google Workspace - Review in the dashboard — Activity and Mosaic show suggested tasks and intent signals
┌─────────────────────────┐
│ Chrome Extension │
│ (TypeScript / React) │
│ - Halo Strip │
│ - Go Live + context │
│ - Chat capture │
│ - Auto-suggestions │
└───────────┬─────────────┘
│ HTTP / WebSocket
┌───────────▼─────────────┐
│ FastAPI Backend │
│ (Python — port 8000) │
│ - /suggest_tasks │
│ - idea_detector │
│ - prompt_builder │
│ - Ask AI + MCP │
└───────────┬─────────────┘
│
┌───────┴────────┐
│ │
┌───▼────┐ ┌─────▼──────┐
│ Gemini │ │ MongoDB │
│ API │ │ Storage │
└────────┘ └────────────┘
┌─────────────────────────┐
│ React Dashboard │
│ (port 3001) │
│ - Activity (tasks) │
│ - Mosaic (intent) │
│ - Session library │
└─────────────────────────┘
- Extension — Captures browsing/chat context, runs the suggestion state machine, surfaces Halo UI
- Processing — FastAPI calls Gemini for task suggestions and prompts; MCP executes Workspace actions
- Storage — MongoDB persists sessions, suggested tasks, and related data (
server/app/db/) - Dashboard — Activity, Mosaic, and library views with WebSocket updates
- Chrome extension: TypeScript, React, Vite (Manifest V3)
- Dashboard: React (port 3001)
- UI: Halo Strip, auto-suggestions, context panel
- FastAPI (Python, port 8000)
- Gemini for task suggestion, Ask AI, and prompt generation
- MCP servers for Google Workspace actions
- MongoDB (local or Atlas) for persistence
- WebSocket for live dashboard / extension sync
Cue does not wait for you to stop and summarize. While Go Live is on, it gates suggestions with cooldowns and intent checks so predictions show up when your focus actually shifts or converges—not as spam.
Searches, site visits, and AI chats feed one context store. That shared context is what makes task suggestions and Antigravity prompts specific to your current thread of work.
Suggested tasks map to real services. Accept a suggestion or use @ commands to draft email, schedule events, or open a build prompt with research already packed in.
Mic sessions can still be transcribed and summarized for a library entry. That path is available; the product center of gravity is the live companion loop.
- Stronger rate limiting and queueing for Gemini calls
- Richer Antigravity / Cursor prompt generation from full research context
- More reliable real-time task status on the dashboard
- Higher-quality task params and confidence ranking
- Node.js (v18+; dashboard expects Node 18)
- Python (3.9+)
- MongoDB (local or Atlas)
- Chrome browser
- Gemini API key
- Google OAuth client (for Workspace actions)
Clone the repository
git clone https://github.com/Siriapps/Cue.git
cd CueInstall dependencies
npm install
cd extension && npm install && cd ..
cd cue && npm install && cd ..
cd server && pip install -r requirements.txtConfigure environment
cp .env.example .env
# Set GEMINI_API_KEY and MONGODB_URI (and Google OAuth as needed)Build the extension
npm run buildLoad extension in Chrome
- Open
chrome://extensions/ - Enable Developer mode
- Load unpacked → select the
extension/distdirectory
Start the backend
npm run start:api
# or: cd server && python -m uvicorn app.main:app --reload --host 0.0.0.0 --port 8000Start the dashboard
npm run start:dashboard
# or: cd cue && npm startFor detailed setup, see SETUP.md.
For run instructions, see RUN.md.
- Capture research across AI chats and docs, then get a paste-ready Antigravity/Cursor prompt when the idea converges
- Turn “I should email the team / block time / open a doc” into suggested Workspace tasks
- Stay in flow while Cue proposes follow-ups from what you’re already reading and discussing
- Execute Gmail / Calendar / Docs actions without leaving the page
- Keep context from tutorials and AI Q&A, then get structured next steps instead of a blank notes page
Built with Gemini, FastAPI, React, MongoDB, and Chrome Extension APIs.
Made by the Cue Team