Skip to content
100% Open Source · Self-Hosted AI Voice Infrastructure

Autonomous phone agents.
Deployed on your terms.

VoiceDesk is the white-label AI voice platform you can inspect, customize, and self-host. Pick a template, design conversational flows, and connect real telephony—with zero vendor lock-in.

MIT Licensed & Free to Self-Host Boots locally with Zero Keys (make up) Cloudflare Python Worker + D1 Serverless Carrier Agnostic (Twilio / Exotel)

Operational Transparency

Inspect the real conversation flow in action.

Select an industry scenario to inspect real-time dialogue turns, automated function calls, and structured telemetry captured during production phone calls.

Dental Appointment & Triage

Healthcare · Aria Clinic · Caller: Maya Chen (+1 (415) 555-0184)

Booked AppointmentDuration: 1m 42s
AI

VoiceDesk Agent

Thank you for calling Northline Dental. I can help schedule cleanings, check emergency slots, or verify insurance. How can I help today?

Maya Chen

Hi, I have a chipped molar and need an urgent appointment early next week.

AI

VoiceDesk Agent

I understand that can be uncomfortable. I have an emergency triage opening with Dr. Taylor on Tuesday at 10:15 AM or Wednesday at 2:00 PM. Which works best?

Maya Chen

Tuesday at 10:15 AM is perfect. Does she accept Delta Dental PPO?

AI

VoiceDesk Agent

Yes, Dr. Taylor is in-network with Delta Dental PPO. I have reserved Tuesday at 10:15 AM for Maya Chen. A confirmation SMS has been sent to your number.

Modular Engineering

Interactive end-to-end architecture.

Click on any stage of the telephony pipeline below to inspect how audio streams, LLM inference, and database persistence coordinate without latency bottlenecks.

Deep Dive: Stage 01

1. Phone Carrier Ingestion

PSTN / SIP

E.164 caller identification (+14155550184) incoming webhook received at edge.

Interactive System Flow

Live Telephony & Agent Execution Lifecycle

Click any stage to inspect protocol

1. Carrier Ingestion

Protocol: Twilio / Exotel SIP Trunk

Avg Latency: < 80ms

Incoming calls from mobile or landline handsets trigger telephony webhooks at the edge.

Inbound caller ID validated via E.164 normalization
Carrier negotiates audio stream directly with WebRTC media engine
SIP warm-transfer targets configured for manual handoff fallback

Enterprise Control

Everything you need to run AI voice in production.

Proprietary voice platforms bill high per-minute margins and keep your prompt logic in closed boxes. VoiceDesk gives you full autonomy.

Zero Database Overhead

Runs locally on SQLite with no external dependencies, or deployable to Cloudflare D1 serverlessly.

Self-Hosted & Private

Your customer data, call recordings, and credentials stay entirely under your own infrastructure control.

Carrier Neutral

Integrates directly with Twilio PSTN or Exotel SIP without locking you into proprietary closed phone networks.

Pluggable AI Models

Works with LiveKit WebRTC, Deepgram, Sarvam AI, OpenAI, and Cartesia voice models out-of-the-box.

Developer-First Architecture

Run locally with zero keys.
Deploy to Cloudflare in one command.

Start developing on your laptop without configuring credentials, databases, or API tokens. When you are ready for cloud hosting, deploy the FastAPI backend as a Cloudflare Python Worker backed by Cloudflare D1 with automatic assets binding.

1

Zero-Key Local Boot

Clone the repository and run make up. The UI and REST endpoints run out-of-the-box.

2

Connect Telephony When Ready

Add an OpenAI or Sarvam key to backend/.env and run make voice for live WebRTC audio.

3

Deploy Serverlessly

Run make deploy-cf to provision Cloudflare D1 and publish the Python Worker.

terminal — bash
Local Quickstart

# 1. Clone repository

git clone https://github.com/opentoolsorg/voicedesk

cd voicedesk


# 2. Boot local dashboard & API (zero keys required)

make up


# 3. Optional: Deploy directly to Cloudflare

make deploy-cf

Dashboard: localhost:3000FastAPI Docs: localhost:8000/docs

Ready to own your voice agent stack?

Download the repository, customize your system prompts, and deploy on your server or Cloudflare in minutes.