Guide

Open-Source and Self-Hosted AI Receptionists

By the BotsRanked Test Team · Last checked: Nov 1, 2026

An open-source AI receptionist is software you run yourself that answers your business phone, talks to callers, and takes a message or books a time. You don't pay a monthly platform fee. You do pay for the phone line, the server, and usually an AI model by the minute. And someone has to set it up and keep it running.

We list 5 projects below. Each card says what the code does, what it needs, and how it connects to a phone number. We haven't tested any of them yet, so every card is marked Not tested. Want a managed service instead? See our AI receptionist rankings.

LobbyStack

Not tested

What it is: an open-source AI receptionist for small businesses, with a dashboard. Also sold as a hosted cloud service.
License: MIT
GitHub stars (GitHub, checked Nov 1, 2026): 46
Last commit (GitHub, checked Nov 1, 2026): Oct 31, 2026
Phone connection: Twilio (Elastic SIP trunk) or Telnyx
What it needs to run: Docker Compose on one server (its guide asks for 2 vCPU, 4 GB RAM) or its Railway template; PostgreSQL and Redis (included); an OpenAI API key; a domain with HTTPS
Receptionist features (per its docs): answers inbound calls, answers from your business info, takes messages, books, moves and cancels appointments in Google Calendar, transfers live calls, saves texts for staff to answer by hand
Note: calendar support is Google Calendar only, per its README. OpenAI handles the live call audio.

AIReceptionist

Not tested

What it is: a self-hosted phone receptionist set up with one YAML file per business.
License: AGPL-3.0
GitHub stars (GitHub, checked Nov 1, 2026): 153
Last commit (GitHub, checked Nov 1, 2026): Sep 16, 2026
Phone connection: SIP, through Twilio, Telnyx or SignalWire trunks, via LiveKit
What it needs to run: Python 3.11+; an OpenAI API key with Realtime access; a LiveKit server (self-hosted or LiveKit Cloud)
Receptionist features (per its README): answers inbound calls, FAQ answers, transfers, takes messages (file, email or webhook), after-hours message, Google Calendar booking, optional recording with a consent notice, more than one business per install
Note: its README says "active development" and warns of breaking changes. Its own cost estimate is about $0.20 to $0.30 per minute of conversation in OpenAI fees. We haven't checked that figure.

Dograh

Not tested

What it is: an open-source voice agent platform with a visual workflow builder. It calls itself an alternative to Vapi and Retell. Also sold as a hosted cloud service.
License: BSD 2-Clause
GitHub stars (GitHub, checked Nov 1, 2026): 6,004
Last commit (GitHub, checked Nov 1, 2026): Oct 31, 2026
Phone connection: Twilio, Vonage, Telnyx, Plivo, Vobiz, Cloudonix, Asterisk ARI
What it needs to run: Docker (one-command setup script); bring your own speech and AI model providers, or use the Dograh defaults
Receptionist features: not a ready-made receptionist. You build one: create an inbound agent, add an FAQ document and a call transfer tool. Booking needs an HTTP tool to an outside calendar, because Dograh has no built-in calendar connector (per its own receptionist guide).
Note: sends anonymous usage data by default; you can turn it off with ENABLE_TELEMETRY=false.

ai-phone

Not tested

What it is: an AI phone agent that runs on your own Mac and can answer and make calls.
License: AGPL-3.0-or-later (bundled PJSIP files are GPLv2 or later)
GitHub stars (GitHub, checked Nov 1, 2026): 373
Last commit (GitHub, checked Nov 1, 2026): Oct 31, 2026
Phone connection: a SIP address, or a Telnyx, Twilio or SignalWire number, or another SIP trunk
What it needs to run: a Mac with Apple Silicon (Intel Macs, Linux and Windows aren't supported); 16 GB RAM recommended for the default local model (8 GB for a smaller one); speech runs on the Mac. The language model can be local, or a Claude Code or Codex subscription, or any OpenAI-compatible API.
Receptionist features: answers inbound calls with your greeting and instructions and keeps a transcript. No built-in booking or message inbox. English only, one call at a time.
Note: new project (first commit Oct 27, 2026). Its README says real-number calls were checked on SignalWire and Telnyx; other checks are still pending.

LiveKit Agents (front desk example)

Not tested

What it is: an open-source framework for voice AI agents. Its repo ships a "front desk" example that books appointments, and a hotel receptionist example.
License: Apache-2.0
GitHub stars (GitHub, checked Nov 1, 2026): 15,037
Last commit (GitHub, checked Nov 1, 2026): Oct 31, 2026 (front desk example last changed Oct 21, 2026)
Phone connection: SIP through LiveKit's telephony stack (open-source LiveKit SIP server or LiveKit Cloud); LiveKit's docs list Twilio, Telnyx, Plivo and Wavix trunks
What it needs to run: Python; a LiveKit server; speech and AI model providers of your choice; a SIP trunk for a real number
Receptionist features: the front desk example lists open slots and books one, optionally through the Cal.com API. It's sample code, not a finished product.
Note: for developers. Expect to write code. Framework is Apache-2.0; its turn-detection models use the separate LiveKit Model License.

What self-hosting means

What it costs

The code is free. Running it isn't. You pay for:

a phone number and call minutes from a carrier like Twilio or Telnyx
a server or computer that stays on (ai-phone runs on a Mac; LobbyStack's guide asks for a 2 vCPU, 4 GB server)
the AI model, usually billed per minute or per token, unless you run a local model like ai-phone can

There's no flat monthly price. Your bill grows with call volume. For managed services, see AI receptionist pricing.

Skills you need

You should be comfortable with Docker or Python, API keys, and a SIP trunk or phone provider console. You'll also handle updates, backups and security fixes. If nobody on your team does that today, a managed service is the safer pick.

Self-hosted vs paid vendors

Self-hosted: you keep call records on your own server, can change anything in the code, and pay providers directly with no platform markup.
Paid vendors: setup takes minutes, someone else fixes outages, and there's support and a fixed plan price.
Either way, the AI provider and carrier still process the call audio, unless the whole stack runs locally.

FAQ

Is there a free open-source AI receptionist?

Yes. LobbyStack (MIT), AIReceptionist (AGPL-3.0), Dograh (BSD 2-Clause) and ai-phone (AGPL-3.0) are free to download and run. You still pay for phone minutes, hosting, and in most setups an AI model.

Can a self-hosted AI receptionist book appointments?

Some can. LobbyStack and AIReceptionist book into Google Calendar. Dograh can book through an HTTP tool you connect to a calendar API. ai-phone has no built-in booking.

Do I need a GPU?

None of these 5 list a GPU as a requirement. Most of these still use cloud AI. LobbyStack and AIReceptionist send call audio to OpenAI. Dograh and LiveKit Agents let you choose your speech and AI providers. ai-phone keeps speech recognition and voice on your Mac, but its language model can run locally or in the cloud.

Which phone providers work with them?

All 5 connect through SIP or a carrier. Twilio and Telnyx are supported by every tool on this page; Dograh also lists Vonage, Plivo and Asterisk.

Has BotsRanked tested these?

Not yet. These cards list facts from each project's own repo and docs. When we test one, it gets the same test calls as paid vendors. See how we test.

Free AI receptionists