Hearthline · AI front desk · home services

An AI front desk for home-service teams, hosted by you.

An open-source AI receptionist for HVAC, plumbing, roofing and solar teams. It answers every inbound call through Vapi, qualifies the lead, books a slot, drafts a quote from a photo, and writes a structured lead into the dashboard. Self-hosted, with your own model and telephony keys encrypted at rest.
Status
Beta
Licence
AGPL-3.0 · commercial licence available
On GitHub
★ 8 · ⑂ 6
inbound calls answered and qualified
24/7
channels: phone, SMS, WhatsApp, email, web
5
UI languages, Japanese included
10
01

Why it exists

Most AI receptionist products are closed services: a monthly fee for each location, a voice the vendor picked, and call data that stays in the vendor's database.

Hearthline turns those three around. The receptionist has your name, your voice and your script. The AI runs on your own API keys, so you pay the provider's price for each call. The leads, calls and quotes live in your own PostgreSQL database.

02

What it does

  • A receptionist with your name and voice

    Set the name, persona, voice and knowledge base for the business. Callers hear your shop answer the phone, at any hour.

  • Your own AI keys

    Anthropic, OpenAI, Vapi and Twilio keys are saved per business, encrypted at rest and masked in the dashboard. You can switch the model provider in settings.

  • One inbox for every channel

    Phone calls, SMS, WhatsApp, email and web chat land on one timeline for each customer.

  • Leads, quotes and tickets

    The agent writes a structured lead into the dashboard and drafts a quote you can edit and print. Inbound messages open support tickets with a threaded reply.

  • Ten dashboard languages

    English, Spanish, German, French, Italian, Portuguese, Dutch, Chinese, Japanese and Arabic, with a right-to-left layout for Arabic.

  • Nine trades to start from

    HVAC, plumbing, windows, roofing, solar, energy renovation, electrical, garage and pest control, each with a knowledge base you can change.

03

How it works

  1. 01

    A customer calls

    Vapi turns speech into text and text into speech, and sends each turn of the conversation to Hearthline.

  2. 02

    The agent decides what to do

    A Django service runs the receptionist loop with Claude or OpenAI. It can qualify the lead, check availability, draft a quote, book an appointment, and send an SMS or an email.

  3. 03

    Everything is written down

    Leads, calls, conversations and quotes are saved in PostgreSQL.

  4. 04

    Your team works from the dashboard

    The Next.js dashboard shows the overview, leads, calls, quotes, customers and support tickets. A test page lets you chat with the receptionist without a phone number.

hearthline / live calls
incoming · +1 (555) 0192
A
Anna
AI receptionist · Apex HVAC
“Hi! This is Anna from Apex HVAC — how can I help you today?”
02:14 · transcribing
~/hearthline/leadwriting → postgres
nameSarah K.
serviceHVAC repair · no heat
address1142 Oak St · zone 3
slotThu 2:00 PM
quote$1,248 · drafted from photo
✓Lead qualified
✓Slot booked · Thu 2:00 PM
✓Photo quote drafted
5 channels · phone · sms · whatsapp · email · webbeta
04

Run it yourself

The whole stack starts with Docker Compose. The demo needs no API keys: every outside service falls back to a stub, so the dashboard still works.

$ git clone https://github.com/codewithmuh/hearthline.git$ cd hearthline$ cp .env.example .env$ docker compose up --build$ docker compose exec backend python manage.py seed_demo --wipe$ docker compose exec backend python manage.py seed_admin
  • Dashboard: http://localhost:3000
  • Django admin: http://localhost:8000/admin
  • REST API: http://localhost:8000/api
  • Add an Anthropic or OpenAI key and the test page becomes a real conversation.
05

Stack

Frontend
Next.js 15React 19TypeScript
Backend
Django 5Django REST Framework
Database
PostgreSQL 16
Voice and messaging
VapiTwilio
AI
Claude (the agent loop)OpenAI (vision, optional provider)
Run and deploy
Docker ComposeVercelCaddy
06

Where it stands

Built

  • Per-business AI and voice keys, encrypted with Fernet
  • The receptionist agent loop with seven tools, on Anthropic or OpenAI
  • Dashboard: overview, leads, calls, quotes you can edit and print, customers, settings and the test simulator
  • Ten dashboard languages
  • Rate limiting on the model endpoint, signed webhook checks, and start-up guards for missing secrets

Not yet

  • Stripe deposits when a quote is accepted
  • Outbound SMS and WhatsApp for sending quotes
  • Several businesses in one deployment (today it is one business per deployment)
  • iPhone and Android companion apps
  • Technician dispatch with routing
07

Questions

Is Hearthline free?
The code is free to host yourself under AGPL-3.0. You pay your own providers (the model, Vapi, Twilio) for what you use. A commercial licence is available for white-labelling or reselling.
Can I try it without API keys?
Yes. The stack runs with a stub for every outside service, and the test page in the dashboard lets you chat with the receptionist. Add a key to make the conversation real.
Which businesses is it for?
Home-service teams: HVAC, plumbing, windows, roofing, solar, energy renovation, electrical, garage and pest control.
Where is the data stored?
In your own PostgreSQL database, on your laptop or your server. Provider keys are encrypted at rest.
Is it ready for production?
It is in beta. One deployment serves one business today, and deposits, outbound messaging and the mobile apps are still in progress.
→Work with me

Want Hearthline running for your team, or something like it built?