The best AI receptionist software depends on whether you want to rent a receptionist or own one. With one location and no technical staff, a ready-made product or Dograh Cloud's visual builder is the fastest start. Teams that want to own the agent can self-host open-source Dograh for free. With local open-weight models, no AI vendor receives the call audio.
Key Takeaways
- Sticker prices mislead, because vendors count different billing units.
- Hosted platforms usually keep a HIPAA business associate agreement for paid or top tiers.
- Self-hosted Dograh with local models keeps AI vendors out. Your carrier still carries calls.
This post is part of our guide to AI receptionists. Here we compare ready-made AI receptionists with platforms you set up yourself. If you would rather own the agent, we show where Dograh, our open-source AI voice agent platform, fits and when owning it is worth it.
Which type fits you
| If you… | Pick |
|---|---|
| Run one location, with nobody technical | A ready-made product (Smith.ai, Rosie, Goodcall…) |
| Want to design your own call flow, without running servers | A hosted builder: Dograh Cloud, Vapi, Retell |
| Need call data to stay on your own servers, or want no per-minute platform fee | Self-hosted Dograh with open-weight models |
| Have engineers who want to write the agent in code | Pipecat or LiveKit Agents |
How we compared the tools
We build Dograh, so every product here, ours included, is checked against the same columns, using each vendor's own pricing page.
We did not run test calls on other vendors' products, so nothing here is a review score. Each price is the list price on the vendor's page, in the vendor's own unit. We did not convert prices into one per-minute rate, because the units do not convert cleanly. A call can last one minute or ten, and one caller can ring back four times in a month.
The table adds two columns that matter once real customer data is on the line. One says whether you can run the software on your own servers. The other says on which plan a business associate agreement (BAA) appears. A BAA is the contract the Health Insurance Portability and Accountability Act (HIPAA) requires before a vendor handles patient information for a healthcare provider.
Where a pricing page says nothing, the table says "not stated". That tells you only that the page is silent on it.
AI receptionist software compared side by side
Every product named in this post is in this table, with the same columns, so you can compare like with like.
| Platform | Open source | Self-host or on-prem | List price | What you get, and what's missing |
|---|---|---|---|---|
| Dograh (open source) | Yes, BSD-2-Clause | Yes | No per-minute platform fee | Build your own on your own servers, with the models you pick. No billing unit of its own, you pay model and phone providers. Self-hosted, no BAA with Dograh needed; model providers may need their own |
| Dograh Cloud | Yes, run as a hosted service | Managed; on-prem on Enterprise | 1¢/min platform fee | Build your own without running servers. Billed per platform minute. HIPAA BAA none published |
| Vapi | No, proprietary | Vapi's cloud | $0.05/min hosting, models and phone extra | Build your own. Billed per platform minute. HIPAA BAA is a $2,000/month add-on |
| Retell AI | No, proprietary | Retell's cloud; dedicated server add-on, or on-prem through Enterprise sales | $0.055/min voice infrastructure, plus models and phone | Build your own. Billed per minute. Self-signed BAA on pay-as-you-go, no extra fee; custom BAA on Enterprise |
| Bland | No, proprietary | Enterprise only (on-prem or VPC) | $0.14/min on Start, models included | Build your own. Billed per minute. HIPAA BAA on Enterprise only |
| Synthflow | No, proprietary | No | Annual enterprise agreement, no per-minute rate | Build your own. Billed by annual contract. HIPAA BAA not stated |
| Pipecat + Pipecat Cloud | Yes, BSD-2-Clause | Yes | Framework free; Cloud $0.01/min active agent (agent-1x), phone $0.018/min extra | Open-source framework. Billed per agent minute. HIPAA BAA not stated |
| LiveKit Agents + LiveKit Cloud | Yes, Apache-2.0 | Yes | Framework free; Cloud $50/mo for 5,000 agent session minutes, then $0.01/min | Open-source framework. Billed monthly plus per minute. HIPAA BAA on Scale ($500/mo) and Enterprise |
| Smith.ai AI Receptionist | No, proprietary | No | $150/mo for 75 calls ($2.00 a call) on Pro | Ready-made. Billed per call. HIPAA BAA not stated |
| Goodcall | No, proprietary | No | $79/mo per agent for 100 unique callers, then $0.50 each | Ready-made. Billed per unique caller. HIPAA BAA not stated |
| Rosie | No, proprietary | No | $49/mo for 250 minutes | Ready-made. Billed per included minute. HIPAA BAA not stated |
| Dialzara | No, proprietary | No | $29/mo for 60 minutes, then $0.48/min | Ready-made. Billed per included minute. HIPAA BAA not stated |
| Upfirst | No, proprietary | No | $24.95/mo for 30 calls, then $1.50 a call | Ready-made. Billed per call. HIPAA BAA on a custom plan |
| Frontdesk (formerly My AI Front Desk) | No, proprietary | No | $99/mo for 200 voice minutes | Ready-made. Billed per included minute, plus credits. HIPAA BAA not stated |
| Reception (ElevenLabs) | No, proprietary | No | $29/mo for 75 credits (1 credit = 1 phone minute) | Ready-made. Billed per included minute. On HIPAA, it "does not currently support HIPAA compliance" |
| Ruby (human reference) | No, not software | No | $720/mo for 200 minutes | Human service. Billed per included minute. "HIPAA-compliant services available" |
| Smith.ai live receptionists (human reference) | No, not software | No | $300/mo for 30 calls, then $11.50 a call | Human service. Billed per call. HIPAA BAA not stated |
Prices are as listed on each vendor's site. Every build-your-own price leaves out the speech and language models and the phone line, except Bland's, which includes the models.
Read the billing unit before the price. Goodcall's $79 counts distinct callers, however long they talk. Dialzara's $29 counts minutes, so a few long calls use up the plan. The cheaper sticker can end up as the bigger bill, depending on how your callers behave.

Ready-made AI receptionists suit a single location
If you run one location and nobody on staff is technical, a ready-made product is the quickest start.
Setup means filling in a form about your business and forwarding your phone number. There are no servers to run and no call flows to design. Building your own only starts to pay off when your calls or your data need control that a rented product cannot give. All prices below are list prices.
Smith.ai's AI Receptionist can hand calls to its North America-based human receptionists, which suits businesses that want a person as backup.
Goodcall charges per unique caller with unlimited minutes, so it suits businesses with long calls.
Rosie texts callers a booking link on its entry plan. Calendar booking and live transfer need its higher Scale plan.
Dialzara has a low-priced entry plan in English and Spanish, but minutes run out fast on long calls.
Upfirst charges per call and offers BAAs only on its Custom plan.
Frontdesk, formerly My AI Front Desk, bundles chat, text messages and a CRM with the phone agent.
Reception, from ElevenLabs: its product page says it "does not currently support HIPAA compliance", so it is not for patient calls.
For scale, human answering costs more per minute. Ruby charges $720 a month for 200 minutes of human receptionists, and Smith.ai's live human receptionists start at $300 a month for 30 calls, then $11.50 a call.
An in-house receptionist earned a median $38,010 a year in May 2025, according to the Bureau of Labor Statistics.
Build-your-own platforms, and where Dograh fits
Build-your-own tools hand you the agent itself, and you take on setting it up and running it.
Dograh: own the agent and run it yourself
Dograh's code is open, so anyone can read it, modify it for their own needs and self-host it for free with Docker Compose.
Its GitHub repository holds the whole platform: telephony, Campaigns for outbound lists, HTTP API tools that read and write your systems during the call, call traces and automatic QA. Agents are built in a visual workflow builder, and can also be built from an AI assistant through Dograh's MCP server. Booking goes through an HTTP API tool to your scheduling system, since there is no built-in calendar connector.
Because Dograh is the orchestration layer that runs the call flow, tools and telephony, the speech and language models are yours to pick, from the platform's managed models to your own provider keys or open-weight models you host.
It works with your own phone setup, including Twilio, Vonage, Plivo, Telnyx and Asterisk ARI. When a call needs a person, Transfer Call hands it over on Twilio, Telnyx or Asterisk ARI, after the agent writes the caller's details to your system, since no summary goes with it by default.
For patient calls, self-host it and make your own models the Organization default on the Models page; then no BAA with Dograh is needed, because it never sees the data, though your model providers and carrier may need their own. Our guide to HIPAA-compliant voice AI covers the full setup.
Self-hosted, Dograh runs 10 calls at once by default, a limit you can raise, with no per-minute platform fee (servers, model keys and phone minutes still cost). Dograh Cloud has a 1¢-a-minute platform fee, with 10 concurrent calls included.
Hosted platforms you build on
Vapi, Retell AI, Bland and Synthflow let you build the agent while they run it on their servers. Our Vapi vs Retell AI comparison puts the first two side by side.
Vapi charges $0.05 a minute for hosting, with models and phone billed on top, and sells HIPAA as a $2,000-a-month add-on. Retell AI charges $0.055 a minute for voice infrastructure, plus models and phone, with a self-signed BAA at no extra fee. Bland charges $0.14 a minute on Start with models included, and keeps its BAA and on-prem or VPC (virtual private cloud) option for Enterprise. Synthflow lists annual enterprise agreements and no self-serve price. Outside Bland's Enterprise on-prem option, the agent lives on the vendor's servers, so leaving means rebuilding it somewhere else. For teams that want a receptionist agent they can run on their own servers, Dograh is the open-source Vapi alternative.
Open-source frameworks you write code on
Pipecat (BSD-2-Clause) and LiveKit Agents (Apache-2.0) are code frameworks. They give you the real-time voice pipeline in Python, and the receptionist itself is yours to write.
Both are free to self-host, and both sell hosting. Pipecat Cloud charges $0.01 per active agent minute on agent-1x, plus $0.018 a minute for phone calls. LiveKit Cloud charges $50 a month for 5,000 agent session minutes on Ship, then $0.01 a minute, and offers a signed BAA from its $500-a-month Scale plan. Dograh is built on a fork of Pipecat, and adds a visual builder and the phone plumbing on top.
Open Source Alternative to Vapi / Retell
Self-hosted voice agent platform — no per-minute fees
dograh-hq/dograh
Star on GitHub
When owning the agent is worth it
Owning the agent makes sense when you want control over the call flow, the costs and where call data goes.
Building pays off in a few cases. One is call data that must not reach a third party. Another is when the voice agent is part of the product you sell, or you run many locations from one setup. If none of those apply to you, rent a ready-made product and revisit the choice when your call volume grows.
Start with cost. Ready-made bills grow with every call or caller. Hosted platforms add a platform fee to every minute on top of the models. Dograh Cloud charges the per-minute platform fee in the table, with 10 concurrent calls included and credits from $5. Self-hosted, the platform fee line is gone, and the rest of the bill is your own servers plus the providers you pick.
Then the data path. HIPAA allows a hosted vendor under a BAA, so self-hosting is a choice the law leaves to you. A BAA delegates obligations to the vendor, but it does not move your own liability. The BAA also tends to cost extra: Vapi's $2,000-a-month add-on, Bland's Enterprise plan, LiveKit's Scale plan and the Custom plan at Upfirst.
For a medical front desk, self-host the platform. No BAA with Dograh is needed, because it never sees the data, though your speech and language providers may need their own BAA and your phone carrier may too. New accounts start on Dograh's managed models. To keep call audio off them, go to the Models page in the dashboard sidebar, set the Organization default to BYOK with your own providers (or local models), and check that no agent has its own model override. Call audio stays fully in-house only when those models run on your own servers too, and with local open-weight models the AI vendor is out of the data path, though your phone carrier still carries the call. Our post on why voice data needs to stay local covers what a recording reveals about a caller.
Join the Dograh Community
Dograh is an OSS alternative to Vapi. Join our Slack community for queries, releases, best practices & community interactions.
What to test in your free trial
Run the same short script on every product you try, over a real phone call, before you compare prices.
Make the route to a person short. In a survey by AnswerConnect, which sells human answering, 85% of 6,000 consumers in the US, UK and Canada said they would rather speak to a real person, and 31% said they would hang up if connected to an AI agent, in its April 2026 survey wave. That is a reason to test the transfer first.
| Test | What to do | Pass rule |
|---|---|---|
| Greeting | Call and listen to the first sentence | It names the business and says what it can help with |
| Interruption | Talk over the agent mid-sentence | It stops and answers the new question |
| After-hours call | Call outside opening hours | It gives the right hours and takes a message or books |
| Wrong department | Ask for something the business does not do | It says so and takes a message or transfers |
| Reschedule | Move an existing booking | The booking moves in your real calendar, with no duplicate |
| Transfer to a person | Ask for a human on a real phone call | Someone answers, or the agent comes back with a set fallback |
| Your monthly bill | Price last month's calls in the vendor's unit | You know the number before you sign |
Exact wording only holds where the platform speaks fixed text. In Dograh, that is the greeting, a pathway's transition speech or a pre-recorded clip. Inside a step the AI generates the words, so put any line that must be said exactly, such as the business name, in the greeting.
Test the transfer over a real phone call, because browser web calls cannot transfer a caller to a phone number. In Dograh, call transfer waits 30 seconds by default, adjustable from 5 to 120. A transfer nobody picks up in that time fails, and the caller is back with the agent, so decide in advance what it does next. It can resume the conversation to take a message, page someone or promise a callback. It can also end the call, or play a voicemail screening message.
If you run one location, pick the ready-made product whose billing unit matches how your callers behave. If you want to own the agent, start with the open-source repository and the build guide linked above.
Glossary
- Billing unit
- The thing a vendor counts to charge you: a call, a unique caller, an included minute or a platform minute. Two plans with the same monthly price can cost very different amounts at your call volume if they count different units.
- Platform fee
- The per-minute charge a voice platform makes for running the agent, separate from the speech and language models and the phone line. Self-hosted Dograh has none, while hosted platforms charge one on every minute.
- Business associate agreement (BAA)
- The HIPAA contract a vendor signs before handling patient information for a healthcare provider. It delegates obligations to the vendor, but the provider keeps its own liability.
- BYOK (bring your own key)
- Using your own account with a model provider inside a voice platform. It moves the contract to you, but the call audio still reaches that provider, so it is not the same as keeping data in-house.

