Vapi makes it quick to get an AI phone agent working. The reasons teams look elsewhere come later — a bill made of several layers, concurrency that has to be bought line by line, a HIPAA add-on priced per month, or a need to run everything on their own servers. This guide compares seven alternatives on what each one publishes: price, limits, and how much of the pipeline you control. Every price below was read on the vendor’s own pricing page on 17 November 2026, and we have marked where we speak from running a platform in production.
First, What Vapi Actually Costs
Vapi charges $0.05 a minute for hosting and passes speech-to-text, the language model and text-to-speech through at the providers’ cost, with no markup. Calls over Vapi’s own telephony are free of a transport fee; Twilio, Vonage and Telnyx bill their own carrier rates. Vapi’s calculator puts 1,000 minutes a month at $82 to $129 with Deepgram, OpenAI and ElevenLabs — roughly $0.08 to $0.13 a minute.
- Vapi’s basic pay-as-you-go option, with no monthly package, includes 4 concurrent calls. The $29-a-month Core package raises that to 10; each extra line is $10 a month.
- HIPAA-eligible data handling, with a Business Associate Agreement, is a $2,000-a-month add-on.
- You choose every provider, which is the main strength — and also why the bill arrives in layers.
The Seven Alternatives at a Glance
| Platform | Published price | What it is | Concurrent calls included |
|---|---|---|---|
| Retell AI | $0.07–$0.31 a minute for voice agents; voice infrastructure $0.055 | Managed platform | 20, then $8 a line a month |
| Bland AI | $0.14 a minute; $0.12 on the $299-a-month Build plan; telephony extra | Managed platform | 10, or 50 on Build |
| ElevenLabs Agents | Plans from $0 to $990 a month with minutes included; $0.08 a minute beyond (Creator plan and up), plus the language model | Managed platform | 4 to 40, by plan |
| Synthflow | Annual enterprise contract; no public price | Managed platform | Agreed per contract |
| LiveKit Agents | Free to self-host; LiveKit Cloud agent minutes $0.01 after the allowance on paid plans | Open-source framework (Apache-2.0) plus optional cloud | 5 or 20 on Cloud plans, more on Scale |
| Pipecat | Free to self-host; Pipecat Cloud $0.01 per active agent minute | Open-source framework (BSD-2-Clause) plus optional cloud | Your servers |
| Dograh | Free to self-host; hosted cloud is usage-based | Open-source platform (BSD-2-Clause) built on Pipecat | Your servers |
Published prices are not a like-for-like comparison: each vendor’s example uses different models, and some include telephony while others do not. Use them to understand how each one bills, then price your own call mix.
Managed Platforms: The Closest Swap for Vapi
These four host everything for you, as Vapi does. You build agents in their dashboard or API and pay per minute. Switching means rebuilding your agents’ prompts, tools and phone-number setup on the new platform.
Retell AI
Retell bills most like Vapi. It quotes $0.07 to $0.31 a minute for voice agents, built from its voice infrastructure at $0.055 a minute plus the language model, the voice and telephony you choose. The notable difference from Vapi is concurrency: 20 simultaneous calls are included before you pay $8 a line a month, and an optional burst setting connects calls above your limit for $0.10 a minute extra instead of rejecting them. Retell describes itself as SOC 2 certified and HIPAA-ready; confirm business associate terms with its sales team.
Bland AI
Bland puts the language model, speech-to-text and voice into one talk-time rate, with no separate model charges: $0.14 a minute with no monthly fee, or $0.12 a minute on the $299-a-month Build plan. Telephony is billed on top, through your own carrier or Bland’s at cost. Transferred time is billed separately at $0.05 or $0.04 a minute. Check the caps before planning an outbound campaign: the Start plan, which has no platform fee, allows 10 concurrent calls and 100 calls a day, and Build allows 50 concurrent and 2,000 a day. Enterprise contracts remove the caps and offer on-premise or private-cloud deployment.
ElevenLabs Agents
ElevenLabs sells agents on monthly plans with call minutes included — from 15 minutes free up to 12,375 minutes on the $990-a-month Business plan — and, from the Creator plan up, $0.08 a minute beyond that. The language model is billed separately by usage, and ElevenLabs adds no fee of its own for telephony; your phone provider bills you directly. Concurrency rises with the plan, from 4 to 40 calls, and optional burst pricing allows up to three times your limit at double the rate. Business associate agreements for HIPAA are offered on Enterprise only. Choose it if you want ElevenLabs’ own voices and a monthly plan with minutes included suits your volume.
Synthflow
Synthflow does not publish self-serve prices. Every deployment starts on an annual enterprise agreement scoped to your call volume, phone system and integrations, with no free plan or trial. The contract can include white label, so BPOs and software platforms can give each client its own workspace under their brand. That suits a BPO; it does not suit a team that wants to try a prototype this week.
Open-Source Frameworks: Own the Pipeline
The other route is to stop paying a platform fee and run the agent yourself. You still pay for speech-to-text, the language model, the voice and the phone line, and you add servers and engineering time. In return you control every part of the pipeline and decide where recordings and transcripts are kept.
LiveKit Agents
LiveKit Agents is an open-source framework (Apache-2.0) for real-time voice agents, with telephony through LiveKit’s SIP stack and a transformer-based model that judges when the caller has finished speaking. You can self-host it, or run it on LiveKit Cloud: the free Build plan includes 1,000 agent session minutes and 5 concurrent sessions, Ship starts at $50 a month with 5,000 minutes and 20 concurrent sessions, and on paid plans minutes beyond the allowance cost $0.01. LiveKit’s own phone-call example — agent session, telephony, language model, speech-to-text, voice and observability — totals $0.0479 a minute. We compared it with Vapi in detail in Vapi vs LiveKit for AI voice agents.
Pipecat
Pipecat is an open-source Python framework (BSD-2-Clause), maintained by Daily, for real-time voice and multimodal agents. It treats the call as a pipeline of steps you assemble yourself, which gives you control over each step and leaves the assembly to your engineers. Daily’s Pipecat Cloud hosts it for $0.01 per active agent minute on the smallest instance, with Daily’s phone dial-in and dial-out at $0.018 a minute.
Dograh
Dograh describes itself as the open-source, self-hostable alternative to Vapi and Retell. It offers a visual workflow builder, bring-your-own speech, language and voice providers, telephony, and a Docker install. It is BSD-2-Clause licensed and built on a fork of Pipecat. A hosted cloud version exists, priced on usage.
What We Run in Production, and What It Costs
Our Indian-language phone agents run on Dograh, self-hosted, with Google Gemini as the language model, Sarvam for speech-to-text, Raya for the voice and Plivo for the phone line. Across 59 live calls between 31 August and 8 September 2026, the provider bill came to ₹1.87 a minute — about $0.02 — with Gemini 62% of it, Plivo 22%, Raya 9% and Sarvam 7%. That excludes the server and our engineering time, and it uses Indian telephony and Indian-language providers, so it is not a price you can compare directly with a US platform. What it does show is where the money goes when there is no platform fee: mostly to the language model.
Self-hosting also means owning the bugs. In the Pipecat version we run, the integration for our speech-to-text provider never marked transcripts as final, so every turn waited for the provider’s published worst case — 1.17 seconds after the caller stopped speaking — even though the transcript was ready after about half a second. On a managed platform that is the vendor’s problem; on your own stack we had to find it in the framework’s code and add a setting to cap the wait. Budget for that kind of work before choosing this route. The full cost breakdown is on our voice AI cost page.
How to Choose
- Stay managed and move to Retell AI if you like Vapi’s model but want more concurrency included.
- Choose Bland AI if you want the model, speech and voice in one per-minute rate rather than picking each provider, and its daily caps fit your volume.
- Choose ElevenLabs Agents if you want ElevenLabs’ own voices and a monthly plan with minutes included suits your volume.
- Choose Synthflow if you run a BPO or software platform, want white label and are ready for an annual contract.
- Choose LiveKit Agents or Pipecat if you have engineers who run infrastructure and want full control, or need recordings and transcripts kept on your own servers.
- Choose Dograh if you want a Vapi-style builder on your own servers without writing the pipeline from scratch.
Questions to Ask Before You Switch
- What is the all-in cost per minute for our call mix — platform, language model, speech-to-text, voice and telephony together?
- How many concurrent calls are included, what does each extra one cost, and what happens to a call above the limit: rejected, or charged more?
- What does HIPAA or a business associate agreement cost, and on which plan?
- Can we bring our own phone numbers or SIP trunk, and does that remove the telephony charge?
- Who holds call recordings and transcripts, for how long, and can we export our agents if we leave?
Frequently Asked Questions
What is the best alternative to Vapi?
For a managed platform that works the way Vapi does, Retell AI is the closest swap, with 20 concurrent calls included. For full control on your own servers, LiveKit Agents or Pipecat; and for a Vapi-style builder you host yourself, Dograh.
Is there an open-source alternative to Vapi?
Yes. LiveKit Agents (Apache-2.0), Pipecat (BSD-2-Clause) and Dograh (BSD-2-Clause) are all open source and can be self-hosted. You pay the speech, language-model, voice and phone providers directly, plus your own servers.
Is Retell AI cheaper than Vapi?
Not on the base fee: Retell’s voice infrastructure is $0.055 a minute against Vapi’s $0.05 hosting, and both charge for the language model and voice on top. Retell can still cost less overall if you need concurrency, because 20 simultaneous calls are included where Vapi includes 4 and charges $10 a month for each extra line.
Is Bland AI cheaper than Vapi?
It depends on your providers. Bland charges $0.12 to $0.14 a minute for the model, speech and voice, plus telephony. Vapi’s own calculator puts a Deepgram, OpenAI and ElevenLabs stack at $0.08 to $0.13 a minute on Vapi’s own telephony; cheaper models bring Vapi lower and premium voices push it higher.
Can I self-host a Vapi alternative?
Yes — LiveKit Agents, Pipecat and Dograh all run on your own servers, and Bland offers on-premise deployment on enterprise contracts. Self-hosting removes the platform fee and lets you decide where recordings and transcripts are stored, at the cost of running and debugging the stack yourself.
There is no single best Vapi alternative — there is the one that matches how you want to pay and how much of the stack you want to own. Retell, Bland, ElevenLabs and Synthflow keep you on a managed platform with different pricing shapes; LiveKit, Pipecat and Dograh move the pipeline onto your servers and replace the platform fee with engineering work. Samcom builds AI voice agents and runs Indian-language agents in production on a self-hosted open-source stack. Tell us your call volume and languages and we will recommend the platform that fits.