PRESENT!

An AI agent that answers your phone calls as you.

Present! picks up incoming phone calls, speaks in your cloned voice, and handles real tasks — searching files, reading documents, checking your calendar, creating events, and sending emails with attachments. You monitor everything in real-time from a live dashboard.

Demo

Call the Twilio number. The AI agent answers as you, has a natural conversation, and executes tasks on your computer — all while you watch the live transcript and approve tool calls from the dashboard.

Live Dashboard: dashboard-ten-rho-71.vercel.app

How It Works

Caller dials your number
        │
        ▼
   Twilio receives call → streams audio via WebSocket
        │
        ▼
   Deepgram STT → real-time speech-to-text (nova-2)
        │
        ▼
   Gemini 2.5 Flash → decides what to say + which tools to use
        │
        ▼
   ElevenLabs TTS → synthesizes response in YOUR cloned voice
        │
        ▼
   Audio streamed back to caller via Twilio
        │
        ▼
   Dashboard shows live transcript, tool calls need your approval

Features

Voice Agent

• Answers phone calls as you — first-person, casual, natural
• Cloned voice via ElevenLabs — sounds like you, not a robot
• Real-time speech — streams TTS audio back to the caller with low latency
• Barge-in support — caller can interrupt the agent mid-sentence
• Adaptive debounce — waits for natural pauses before responding

Tool Execution

The agent can perform real actions on your computer during a call:

Tool	What it does
list_files	Searches Desktop/Documents/Downloads for files
open_file	Opens and reads .docx, .txt, .csv, .json, .md files
check_calendar	Queries Apple Calendar for events
create_event	Creates calendar events via .ics files
send_email	Sends emails with attachments via Outlook/Mail.app
browse_url	Opens URLs in a headed browser

Dashboard

• Live transcript — see the conversation as it happens, with timestamps
• Tool approval gate — Claude Code-style Allow/Deny before any tool executes
• Call duration timer — live elapsed time counter
• Fullscreen chat mode — expand to full viewport for mobile use
• Past sessions — browse previous calls with expandable transcripts (persisted in Supabase)
• Voice cloning — record or upload audio to clone your voice from the dashboard
• Copy transcript — one-click copy of the full conversation
• Confirmation dialogs — safety net before ending calls
• Toast notifications — feedback for actions
• Mobile-friendly — deployed on Vercel, accessible from your phone

Tech Stack

Layer	Technology
Phone	Twilio (Voice API + Media Streams WebSocket)
Speech-to-Text	Deepgram (nova-2, real-time streaming)
LLM	Google Gemini 2.5 Flash (multi-round tool calling)
Text-to-Speech	ElevenLabs (turbo v2.5, instant voice cloning)
Tool Execution	Playwright (headed Chromium for browser automation)
Server	Node.js, Express, Socket.IO
Dashboard	Next.js 16, React 19, Tailwind CSS 4
Database	Supabase (PostgreSQL — session history)
Hosting	Vercel (dashboard), ngrok (server tunnel)

Architecture

┌─────────────────────────────────────────────────────┐
│                    Dashboard (Next.js)               │
│  Live transcript │ Tool approval │ Voice cloning     │
│  Past sessions   │ Fullscreen    │ Connection status │
└────────────┬────────────────────────────────────────┘
             │ Socket.IO (real-time)
             ▼
┌─────────────────────────────────────────────────────┐
│                    Server (Express)                   │
│                                                       │
│  ┌──────────┐  ┌──────────────┐  ┌────────────────┐ │
│  │ Twilio   │  │ Orchestrator │  │ Browser Agent  │ │
│  │ Media WS │→ │ (Gemini)     │→ │ (Playwright)   │ │
│  └──────────┘  └──────────────┘  └────────────────┘ │
│       │              │                    │           │
│       ▼              ▼                    ▼           │
│  ┌──────────┐  ┌──────────────┐  ┌────────────────┐ │
│  │ Deepgram │  │ ElevenLabs   │  │ Files, Calendar│ │
│  │ STT      │  │ TTS          │  │ Email, Browser │ │
│  └──────────┘  └──────────────┘  └────────────────┘ │
│                                                       │
│  ┌──────────────────────────────────────────────────┐│
│  │ Supabase — session history + transcripts         ││
│  └──────────────────────────────────────────────────┘│
└─────────────────────────────────────────────────────┘

Setup

Prerequisites

• Node.js 24+
• Twilio account with a phone number
• Deepgram API key
• Google Gemini API key
• ElevenLabs API key
• Supabase project
• ngrok (for exposing the server to Twilio)

Environment Variables

Copy .env.example to .env and fill in:

TWILIO_ACCOUNT_SID=
TWILIO_AUTH_TOKEN=
TWILIO_PHONE_NUMBER=
DEEPGRAM_API_KEY=
GEMINI_API_KEY=
ELEVENLABS_API_KEY=
ELEVENLABS_VOICE_ID=
SERVER_URL=           # ngrok host (e.g. abc.ngrok-free.app)
SUPABASE_URL=
SUPABASE_ANON_KEY=

Running

# Terminal 1 — expose server to Twilio
ngrok http 3001

# Terminal 2 — start server
cd server && npm install && npm run dev

# Terminal 3 — start dashboard
cd dashboard && npm install && npm run dev

Set your Twilio Voice webhook to https://<ngrok-host>/api/twilio/voice.

Open http://localhost:3000 to see the dashboard.

Team

Built by Lucas Tran (@lukietee)