The best AI phone agents are picked by the job on the line, such as after-hours answering, bookings or calling a list. Vapi, Retell AI and Bland are hosted platforms, and Smith.ai and PolyAI run it for you. For teams that want to own the platform, Dograh is open source, and you can self-host it or use Dograh Cloud.
Key Takeaways
- Pick the phone agent by the job on the line, then by who runs it.
- Smith.ai bills per call, while Vapi, Retell AI and Bland bill per minute.
- Open-source Dograh lets a regulated line run on your own servers.
An AI phone agent answers or places calls for a business, and the work differs from one phone line to the next. We start with five jobs a phone line holds, match each to the products that fit, then talk about how prices compare and where open-source Dograh fits.
What counts as an AI phone agent
An AI phone agent is software that holds a real phone conversation for a business. It listens, answers, asks questions, looks things up and hands the call to a person when it should. Most teams use it for service calls first. In RingCentral's agentic voice AI report, 57% of leaders said they had already deployed voice AI agents, with customer support named by 51% as the top use, then scheduling at 18% and call routing at 15%.
Who runs the agent splits the market three ways. On a hosted platform your team builds the agent and the vendor runs it. With a managed service the vendor builds it and runs it for you. With open source you run the whole platform on your own servers or on its vendor's cloud. We rank platforms by who runs them, with a 10,000-minute price table, in our list of best AI voice agent platforms, and services where a person answers every call sit in our list of AI phone answering services.

Five jobs on a phone line and what fits
Each job asks something different of the agent, so the right product changes with it. The same report found 70% of enterprises accept voice AI only with strong guardrails, verified accuracy and a handoff to a person, against 41% of small businesses. That handoff shows up in three of the five jobs below.
| Job on the line | What the agent must do | Type that fits | Picks |
|---|---|---|---|
| Answer the main line after hours | Take every call, answer common questions, take a message | Managed service or hosted platform | Smith.ai, Retell AI, Dograh |
| Book from live slots | Read open times from your system and write the booking back | Hosted or open-source platform with API tools | Retell AI, Vapi, Dograh |
| Call a list | Dial contacts with retries, set calling hours, record each outcome | Platform with campaigns | Bland, Dograh, Retell AI |
| Screen callers and hand them to staff | Ask a few questions, then pass the caller to the right person | Platform with a transfer that waits for an answer | Dograh, Retell AI, Synthflow |
| A regulated line on your own servers | Keep audio, transcripts and recordings inside your network | Open-source platform you self-host | Dograh |
The last row is the narrowest. A closed product can run on-prem only under an enterprise deal, while open code lets your own team run and inspect it.
AI phone agents compared
The table puts the published billing unit, the transfer behavior and where each product runs side by side. Every row except Dograh and the frameworks is closed code. For teams that would rather write the agent in code, our list of voice AI frameworks covers Pipecat, LiveKit Agents and the newer agent SDKs.
| Product | Who runs it | How it bills | Hands calls to staff | Where it runs | Code |
|---|---|---|---|---|---|
| Dograh | Your team, self-hosted or on Dograh Cloud | Self-hosted, no per-minute platform fee. Cloud, 1¢/min platform fee. Models and phone extra | Rings the destination first, caller stays with the agent if nobody answers | Your servers, Dograh Cloud, on-prem on Enterprise | Open source, BSD 2-Clause |
| Vapi | Your developers, on Vapi's cloud | $0.05/min hosting plus models and phone | Blind by default, spoken summary modes on Twilio only | Vapi's cloud | Closed |
| Retell AI | Your team, on Retell's cloud | $0.055/min voice infrastructure, $0.07 to $0.31/min all-in | Cold, warm with a private whisper, or an AI that briefs the person | Retell's cloud, on-prem through sales | Closed |
| Bland AI | Your team, on Bland's cloud | Start $0.14/min with models included | Standard transfer, briefed transfer on Enterprise | Bland's cloud, on-prem or VPC on Enterprise | Closed |
| ElevenLabs Agents | Your team, on ElevenLabs' cloud | Plans with minutes included, $0.08 per extra minute, LLM extra | Warm with a summary on Twilio, SIP REFER | ElevenLabs' cloud | Closed |
| Smith.ai AI Receptionist | Smith.ai | Per call, Pro from $150/mo | Live North America-based receptionists | Smith.ai's service | Closed |
| PolyAI | PolyAI and its partners | Per minute, by quote | SIP REFER or INVITE to your contact center | PolyAI's cloud | Closed |
| Synthflow | Your team or an agency, on an annual contract | Annual enterprise agreement | Cold, or warm after it detects a person | Synthflow's cloud | Closed |
| Pipecat, LiveKit Agents | Your developers | Free to self-host | Built in your code | Your servers | Open source |
Dograh: an open-source phone agent you can run
Dograh suits the jobs where you want the phone agent on your own terms, a regulated line that must stay on your servers or a team that wants to change every part of the call. The whole platform is open source under BSD 2-Clause, and you can self-host it with Docker Compose on a server with 8 GB of RAM and 4 vCPUs. New agents start in the Agent Builder, where you describe the call, approve the requirements card and let it build, with a canvas for hand edits.
It runs on your own carrier account, on Twilio, Telnyx, Plivo, Vonage, Exotel, Vobiz, Cloudonix or Asterisk ARI, so the phone line and its numbers stay yours. For the screening job, Transfer Call works on Twilio, Telnyx, Plivo, Cloudonix and Asterisk ARI. Dograh calls the destination first and connects the caller only when someone answers. When the person is busy, never picks up or the transfer fails, the agent keeps the caller and carries on. On Exotel, Vobiz and Vonage numbers, the agent books a callback request through an HTTP API tool when the caller needs a person.
For calling a list, Campaigns dial from a contact list with retries and calling hours set in one time zone per campaign. You pick the speech-to-text, language model and voice for each agent, from Dograh's models, your own keys or open-weight models on your own servers.
Where it runs sets the cost. Running Dograh yourself means no per-minute platform fee. You still pay for servers, model keys and phone minutes, and your team looks after the database and upgrades. A self-hosted install runs 10 calls at once by default, which you can raise. Support comes from the open-source community, and Enterprise adds custom SLAs and on-prem deployment. If you would rather not host it, Dograh Cloud runs it for a 1 cent a minute platform fee and includes 10 calls at once, with more on request. A platform still needs someone to build the agent, so a small office with nobody to do that may be better served by a managed service.
Open Source Alternative to Vapi / Retell
Self-hosted voice agent platform — no per-minute fees
dograh-hq/dograh
Star on GitHub
Hosted platforms: Vapi, Retell AI, Bland, ElevenLabs
Hosted platforms let your team build the agent while the vendor runs the servers, and all four publish rates. Vapi charges $0.05 a minute for hosting, with speech, the language model, the voice and the phone line billed at cost on top. Without a plan you get 4 calls at once, and Core adds 10 for $29 a month. It fits developers who want to wire every part in code.
Retell AI charges $0.055 a minute for voice infrastructure and $0.015 a minute for telephony on its numbers, with no charge for your own SIP trunk. Its own all-in range runs from $0.07 to $0.31 a minute, and 20 calls at once come included. Its transfer options are the widest of the hosted group, from a cold handoff to a separate AI that briefs the person who answers and returns the caller if they cannot take it.
Bland's Start plan costs $0.14 a minute with the language model, speech-to-text and voice included, capped at 10 calls at once and 100 calls a day. Build costs $299 a month plus $0.12 a minute and lifts the cap to 2,000 calls a day, which suits the call-a-list job. ElevenLabs Agents sells plans with minutes included, from Pro at $99 for 1,238 minutes to Scale at $299 for 3,738, and bills the language model on top.
How each one hands a caller to a person differs more than the rates do. Vapi's spoken summary modes work on Twilio only, and on Twilio, Telnyx and Vonage an unanswered blind transfer ends the call. Our list of voice AI with human handoff compares that behavior vendor by vendor.
Managed services that build the agent for you
A managed service suits an office with nobody to build or maintain an agent. Smith.ai's AI Receptionist bills per call, with a free plan of 25 calls and Pro from $150 a month for 75 calls, and a complex call can go to a live North America-based receptionist at any hour. Our guide to voice AI for small businesses compares it with other done-for-you options.
PolyAI prices its managed agents through its sales team, per minute, and sells its Dialog-RSN-1 model by API at 2 cents a connected minute. PolyAI builds the agent with you and hands calls to your contact center over SIP. Synthflow sells every new deployment on an annual enterprise agreement scoped to your call volume, with no free plan or trial. Both run on their own clouds and keep the code closed, so recordings and transcripts sit with the vendor.
What 3,000 minutes a month costs
A worked example makes the billing units concrete. Take 1,000 calls of three minutes each, or 3,000 minutes, and price only what each vendor publishes for them.
Vapi's hosting comes to 3,000 x $0.05 = $150, before models and the phone line. Retell AI's voice infrastructure is 3,000 x $0.055 = $165, and its own all-in range gives about $210 to $930. Bland Start is 3,000 x $0.14 = $420 with models included and the phone line billed separately. ElevenLabs Scale at $299 covers 3,738 minutes, with the language model on top.
Smith.ai counts calls instead of minutes, so 1,000 calls sit in its Enterprise range, which starts at $500 a month and is quoted at that volume. Dograh Cloud's platform fee is 3,000 x $0.01 = $30, with your own models and carrier added, and a self-hosted install has no platform fee at all. PolyAI and Synthflow set their price in the quote. The rows count different things, so compare each one with the models and phone minutes added.
Join the Dograh Community
Dograh is an OSS alternative to Vapi. Join our Slack community for queries, releases, best practices & community interactions.
Picking an AI phone agent by job
Start from the job on the line. For after-hours answering with nobody to build an agent, a managed service such as Smith.ai is the quickest route. For booking and calling lists, a hosted platform or Dograh works, and the choice comes down to who runs the servers and how much you want to change. For screening callers, pick a product whose transfer rings the person first and keeps the caller when nobody answers.
For a regulated line, or any team that wants the agent and its call data on its own servers, start with Dograh, with an Enterprise contract behind it when you need on-prem. Whatever you shortlist, test it over a real phone call before launch. An agent that sounds sharp in a demo can mishear on the line, because most phone calls still reach the agent as 8 kHz audio, while a browser demo runs at 16 kHz.
Glossary
- Platform fee
- The charge a voice agent platform makes for running the call, separate from the models and the phone line. Vapi charges $0.05 a minute for hosting, and Dograh Cloud charges a 1¢-a-minute platform fee.
- Per-call billing
- Pricing by the number of calls handled, whatever their length, as Smith.ai does. Long calls cost the same as short ones, so it suits lines with longer conversations.
- Concurrent calls
- How many calls an agent can hold at the same moment. Vapi allows 4 without a plan, Retell AI includes 20, and Dograh Cloud includes 10.
- Managed service
- A phone agent the vendor builds and runs for you, such as Smith.ai or PolyAI. You trade control over the agent for not having to build it.

