Voice AI agent platforms compared (2026)

We read the official sites, docs, pricing pages and GitHub repositories of 11 voice AI agent platforms on 2026-11-01. Bland AI, Deepgram Voice Agent API, ElevenLabs Agents Platform, LiveKit Agents, Retell AI and Vapi state HIPAA compliance or a BAA, and Bland AI, Deepgram Voice Agent API, ElevenLabs Agents Platform, PolyAI and Vapi document EU data hosting. "Not stated" means we found no statement on those pages, not that the feature is missing.

Last checked 2026-11-01. Tools are listed alphabetically. This is not a ranking and we did not test the tools ourselves. Enterprise prices, customer counts and market share are not included because we could not source them. Nothing here is a paid placement.

Comparison table

Platform	Type	Open source / license	Self-host	HIPAA / BAA	EU data	Pricing as listed 2026-11-01
Bland AI	managed voice agent platform	Not stated	Yes	Yes	Yes	Start $0.14/min with $0 platform fee and no card required; Build $0.12/min plus a $299/month platform fee; Enterprise is custom, contracted to your volume.
Deepgram Voice Agent API	voice agent API	Not stated	Not stated	Yes	Yes	pay as you go from $0.050 to $0.163 per minute depending on the configuration (Standard $0.075, Advanced $0.163, Custom with your own LLM and TTS $0.050). One rate covers speech-to-text, a managed language model and text-to-speech. A free $200 credit with no card is listed. Growth is prepaid from $4,000+ per year. Enterprise is custom.
ElevenLabs Agents Platform	managed voice agent platform	Not stated	Not stated	Yes	Yes	the Free plan includes 15 call minutes; additional call minutes cost $0.08 per minute on every plan, and burst pricing is $0.160 per minute. Included minutes by plan: Starter 75, Creator 275, Pro 1,238, Scale 3,738, Business 12,375. Enterprise is custom.
LiveKit Agents	open-source voice agent framework	Yes: Apache-2.0	Yes	Yes	Not stated	As listed on 2026-11-01 for LiveKit Cloud: Build is $0/month with no card and 1,000 agent session minutes included; Ship includes 5,000 minutes and Scale 50,000, then $0.01 per minute; Enterprise is custom. Self-hosting the framework is free under the license; model and telephony providers bill separately.
Pipecat	open-source voice agent framework	Yes: BSD-2-Clause	Yes	Not stated	Not stated	Not stated
PolyAI	enterprise voice assistant (managed service)	Not stated	Not stated	Not stated	Yes	the platform page lists no prices and says ongoing use is priced per minute, with a "Talk to our team" form. The separate Dialog-RSN-1 API is listed at "2¢ per min", billed per second of connected time, excluding telephony and orchestration, and sales is needed only above 10 concurrent calls.
Retell AI	managed voice agent platform	Not stated	Not stated	Yes	Not listed	pay as you go shows $0.07-$0.31 / min for AI voice agents with $10 in free credits; Retell Voice Infra $0.055/minute; telephony (Twilio, US) $0.015/min. Enterprise is custom.
Synthflow	managed voice agent platform	Not stated	Not stated	Not stated	Not stated	no published per-minute rate. The page says every new deployment starts on an Enterprise contract scoped to call volume, concurrency, telephony and integrations, and that there is no free plan or trial.
Twilio ConversationRelay	voice agent API	Not stated	Not stated	Not stated	Not stated	$0.07 / minute, with the note "Voice costs are calculated separately." The page is marked "Pricing current as of October 2026" and says custom or high-volume pricing needs sales.
Vapi	voice agent API	Not stated	Not stated	Yes	Yes	Vapi hosting is $0.05/min on the pay-as-you-go package, plus provider costs shown as ranges (transcriber $0.0095-$0.0099/min, language model $0.0077-$0.0452/min, voice $0.0146-$0.0238/min) and telephony charged by the provider (for example Twilio inbound $0.008/min). $5 in free credits is listed. Premier is custom, direct sales only.
Voiceflow	managed voice agent platform	Not stated	Not stated	Not stated	Not stated	the pricing page shows no prices. It offers a free trial with no credit card for agencies and partners, "Transparent, usage-based billing", and a "Request pricing" button for businesses.

"Yes" for HIPAA / BAA means the vendor's own page states HIPAA compliance or a BAA, often on a specific plan (see the card). "Yes" for EU data means the vendor documents EU hosting or an EU endpoint for this product. "Not listed" means the vendor's docs say it does not offer it. "Not stated" means we found no clear statement. "Yes" for self-host includes Enterprise-only private or on-premises deployment, which the cards say. Prices change often, so confirm on the vendor's pricing page.

Tool by tool

Bland AI

Platform for AI phone agents with self-serve plans and an Enterprise tier that adds BAA, data residency and VPC or on-prem deployment.

Type: managed voice agent platform

Suits: Teams that want a managed platform and are prepared to buy Enterprise for BAA, regional pinning or private deployment.

Compliance: The pricing FAQ says "HIPAA compliant with a signed BAA", and the pricing table lists BAA on Enterprise only. The pricing FAQ also states SOC 2 Type I and Type II and GDPR compliance with a DPA. The security page says calls can be pinned to a region you specify and that "EU data stays on EU GPUs", and it notes that wrappers route inference through US-based provider clouds. The pricing table lists data residency on Enterprise only.

Pricing: As listed on 2026-11-01: Start $0.14/min with $0 platform fee and no card required; Build $0.12/min plus a $299/month platform fee; Enterprise is custom, contracted to your volume.

Notes: Self-hosting (on-prem or VPC) is listed as available on Enterprise only. The security page also lists air-gapped deployment.

Sources: bland.ai/pricing, bland.ai/security

Deepgram Voice Agent API

Deepgram says its Voice Agent API "handles the full speech pipeline (listening, thinking, and speaking)".

Type: voice agent API

Suits: Developers who want one API for the speech pipeline and will build the phone, call-flow and business logic around it.

Compliance: The pricing page says "We sign Business Associate Agreements (BAA) for Enterprise customers" handling ePHI. Deepgram's regional endpoints docs list the Voice Agent API (/v1/agent/converse) on the EU endpoint, with the WebSocket URL wss://api.eu.deepgram.com/v1/agent/converse, and say requests to the EU endpoint are never routed outside the EU.

Pricing: As listed on 2026-11-01: pay as you go from $0.050 to $0.163 per minute depending on the configuration (Standard $0.075, Advanced $0.163, Custom with your own LLM and TTS $0.050). One rate covers speech-to-text, a managed language model and text-to-speech. A free $200 credit with no card is listed. Growth is prepaid from $4,000+ per year. Enterprise is custom.

Notes: The pricing page describes self-hosted containers in your VPC or on-prem for Enterprise, but does not say they cover the Voice Agent API, so self-host is left unstated. We did not find a statement that a third-party LLM you configure stays in the EU.

Sources: deepgram.com/pricing, developers.deepgram.com/reference/regional-endpoint, developers.deepgram.com/docs/voice-agent

ElevenLabs Agents Platform

ElevenLabs describes its agents platform as a way to build, deploy and monitor voice AI agents that hold natural conversations and take actions.

Type: managed voice agent platform

Suits: Teams that want a managed agent with ElevenLabs voices and a native Twilio or SIP trunk connection.

Compliance: The agents HIPAA docs say ElevenLabs Agents is HIPAA-eligible and that BAAs are available only on Enterprise. Handling PHI needs both a BAA and Zero Retention Mode. The data residency docs say storage in the EU, India and Singapore is an Enterprise feature and that customer data is stored in the US by default. The agents page lists "Data residency in US, EU, and India" under its enterprise section.

Pricing: As listed on 2026-11-01: the Free plan includes 15 call minutes; additional call minutes cost $0.08 per minute on every plan, and burst pricing is $0.160 per minute. Included minutes by plan: Starter 75, Creator 275, Pro 1,238, Scale 3,738, Business 12,375. Enterprise is custom.

Notes: The agents page lists "On-prem options" without saying what they cover, so self-host is left unstated. The data residency docs do not confirm availability of every Agents Platform feature in each region.

Sources: elevenlabs.io/agents, elevenlabs.io/pricing/agents, elevenlabs.io/docs/eleven-agents/legal/hi, elevenlabs.io/docs/overview/administratio, elevenlabs.io/docs/eleven-agents/overview

LiveKit Agents

The GitHub repo describes it as "A framework for building realtime voice AI agents", with a hosted LiveKit Cloud as an option.

Type: open-source voice agent framework

Suits: Engineering teams that want to own the agent code and run it on their own servers or on LiveKit Cloud.

Compliance: The LiveKit Cloud pricing table lists "Signed BAA" under HIPAA compliance. The table is hard to read and does not say which tier includes it, so confirm the plan with LiveKit. The pricing page lists region pinning, which restricts which data center regions may route or process media streams. We found no EU data residency statement for the agents product.

Pricing: As listed on 2026-11-01 for LiveKit Cloud: Build is $0/month with no card and 1,000 agent session minutes included; Ship includes 5,000 minutes and Scale 50,000, then $0.01 per minute; Enterprise is custom. Self-hosting the framework is free under the license; model and telephony providers bill separately.

Notes: The repo says the stack is open source and can run on your own servers. The turn detection models are under the separate LiveKit Model License, not Apache-2.0. The pricing page says on-premise or private deployments are available through sales.

Sources: github.com/livekit/agents, livekit.com/pricing

Pipecat

The repo describes Pipecat as an "Open Source framework for voice agents, multimodal apps, and realtime AI."

Type: open-source voice agent framework

Suits: Python developers who want to assemble speech, language and telephony components themselves.

Notes: The site says it runs wherever Python runs and lists WebRTC, SIP and PSTN transports. It also describes Pipecat Cloud, a managed hosting service, and Pipecat Enterprise, which deploys Pipecat Cloud in your own AWS, Azure or GCP environment through sales. The site gives no prices and we found no HIPAA or EU statement.

Sources: pipecat.ai, github.com/pipecat-ai/pipecat

PolyAI

PolyAI says it "builds dialog agents that handle your most critical enterprise conversations, resolving them without escalation."

Type: enterprise voice assistant (managed service)

Suits: Larger organizations that want a managed, vendor-run voice assistant rather than a developer toolkit.

Compliance: The home page lists "SOC 2+ Type 2, ISO 27001, PCI DSS, GDPR" and, under data residency, "US, UK, EU, APAC" with no cross-region transfer. HIPAA is not mentioned on that page.

Pricing: As listed on 2026-11-01: the platform page lists no prices and says ongoing use is priced per minute, with a "Talk to our team" form. The separate Dialog-RSN-1 API is listed at "2¢ per min", billed per second of connected time, excluding telephony and orchestration, and sales is needed only above 10 concurrent calls.

Notes: The platform is sold as a service. The pricing page does not say a sales call is mandatory for the platform, but gives no price for it.

Sources: poly.ai, poly.ai/pricing

Retell AI

Platform for AI voice agents, with pay as you go and Enterprise plans.

Type: managed voice agent platform

Suits: US-based teams that need a BAA and want a managed platform, and do not require EU-hosted processing.

Compliance: The compliance docs say a signed BAA is required before sending PHI, that the BAA can be self-signed with no additional fee, and that Retell is SOC 2 Type 1 and Type 2 certified. GDPR compliance is described through AWS infrastructure and a DPA with SCCs. The same page says "we do not currently operate services within the European Union."

Pricing: As listed on 2026-11-01: pay as you go shows $0.07-$0.31 / min for AI voice agents with $10 in free credits; Retell Voice Infra $0.055/minute; telephony (Twilio, US) $0.015/min. Enterprise is custom.

Notes: The pricing page also says "SOC 2 certified, HIPAA-ready" for Enterprise. Its calculator separates the voice agent rate from telephony.

Sources: retellai.com/pricing, docs.retellai.com/general/compliance

Synthflow

Platform for AI voice agents; its pricing page describes an annual enterprise agreement.

Type: managed voice agent platform

Suits: Organizations ready to go through a sales-led annual contract for a managed platform.

Compliance: The pricing page says "Synthflow maintains SOC 2 Type 2, ISO/IEC 27001, HIPAA, and GDPR coverage." It does not mention a BAA. The page shows an EU hosting badge but gives no detail on where data is stored or processed, so HIPAA and EU data are left unstated here.

Pricing: As listed on 2026-11-01: no published per-minute rate. The page says every new deployment starts on an Enterprise contract scoped to call volume, concurrency, telephony and integrations, and that there is no free plan or trial.

Notes: Pricing is not published, so cost cannot be compared per minute with the other tools here.

Sources: synthflow.ai/pricing

Twilio ConversationRelay

Twilio's conversational AI voice product; its page describes it as "Human-friendly voice AI".

Type: voice agent API

Suits: Teams that already run their phone numbers on Twilio.

Compliance: The product page says ConversationRelay supports "HIPAA-eligible architectures". It does not mention a BAA, and it lists no regions beyond "Global connectivity". Confirm HIPAA terms and regions with Twilio before relying on it.

Pricing: As listed on 2026-11-01: $0.07 / minute, with the note "Voice costs are calculated separately." The page is marked "Pricing current as of October 2026" and says custom or high-volume pricing needs sales.

Notes: The relay rate is charged on top of Twilio voice charges. The page shows a free trial option on Twilio.

Sources: twilio.com/en-us/products/conversation, twilio.com/en-us/products/conversation

Vapi

Developer platform for building voice agents on top of the transcriber, language model and voice providers you choose.

Type: voice agent API

Suits: Developers who want to choose each model and telephony provider and pay per component.

Compliance: The HIPAA docs say HIPAA is available "as a paid compliance add-on with Usage only or any Success Package" and that Vapi emails the BAA for signature. The pricing page lists the add-on at $2,000/month. The EU region docs say it is "available only with the Premier Success Package" and that Vapi's EU region lets you run agents hosted in the European Union. The docs point to a separate data flow page for exactly what is processed where.

Pricing: As listed on 2026-11-01: Vapi hosting is $0.05/min on the pay-as-you-go package, plus provider costs shown as ranges (transcriber $0.0095-$0.0099/min, language model $0.0077-$0.0452/min, voice $0.0146-$0.0238/min) and telephony charged by the provider (for example Twilio inbound $0.008/min). $5 in free credits is listed. Premier is custom, direct sales only.

Notes: Total cost per minute is the sum of hosting and the providers you select. A US and an EU organization are separate regions with separate endpoints.

Sources: vapi.ai/pricing, docs.vapi.ai/security-and-privacy/hipaa, docs.vapi.ai/security-and-privacy/eu-reg

Voiceflow

Platform for building and deploying conversational agents across voice and chat.

Type: managed voice agent platform

Suits: Teams that are willing to request pricing and want a hosted platform for voice and chat agents.

Compliance: The security page lists HIPAA, SOC 2 Type II and GDPR as compliance frameworks and links to further compliance pages. It does not mention a BAA or EU data residency, so HIPAA and EU data are left unstated here.

Pricing: As listed on 2026-11-01: the pricing page shows no prices. It offers a free trial with no credit card for agencies and partners, "Transparent, usage-based billing", and a "Request pricing" button for businesses.

Notes: The pricing page says production environments are all hosted on Voiceflow. We did not find phone-call pricing or limits on the pages we read.

Sources: voiceflow.com/pricing, voiceflow.com/security

How to choose

1. Start from your compliance requirement

If you handle patient data, a signed BAA is the starting point, not a badge on a website. Bland AI, Deepgram Voice Agent API, ElevenLabs Agents Platform, LiveKit Agents, Retell AI and Vapi state HIPAA compliance or a BAA on the pages we read, and the plan that includes it differs, so ask which plan and what it costs. If you need data kept in the EU, Bland AI, Deepgram Voice Agent API, ElevenLabs Agents Platform, PolyAI and Vapi document it. Check what stays in the region: the models you connect may process data elsewhere.

2. Decide managed platform or build it yourself

A managed platform gives you a dashboard and a working phone agent sooner. An API or open-source framework such as LiveKit Agents and Pipecat gives you control over each step and needs engineering time to build, host and monitor. Self-hosting also moves the compliance question to your own infrastructure.

3. Check the telephony you already have

If your numbers and call routing are already on Twilio or a SIP trunk, check how each platform connects to it and who bills the minutes. Twilio ConversationRelay is Twilio's own product, and ElevenLabs lists a native Twilio integration and SIP trunking. Vapi and Retell AI list telephony as a separate line on their pricing pages.

4. Work out cost per minute at your call volume

Per-minute prices on these pages are often a floor or a range, and telephony, speech and model costs can be billed separately. Multiply the listed rate by your monthly minutes, then add the other lines. Pipecat and Voiceflow publish no per-minute price on the pages we read, so you will need a quote.

5. Test latency and interruptions on your own calls

We did not compare latency or how well each agent handles callers who interrupt, because we did not test the tools and could not confirm comparable figures on the vendors' own pages. Run a trial on real call recordings and measure it yourself.

Also see Vapi vs Retell AI for a side-by-side of two of these platforms, or the full tools hub.

Need someone to build it?

If you would rather hand this to a specialist, browse the voice AI agencies in our directory, or the lists for the US, the UK and Australia. See agencies for voice agents, compare Vapi and Retell AI, or get matched with an agency. Our listings do not say which platform an agency uses, so ask about it in your brief, along with who signs the BAA.

Sources

All fetched on 2026-11-01.

Questions

Which voice AI platforms offer HIPAA or a BAA?

On the vendor pages we read, Bland AI, Deepgram Voice Agent API, ElevenLabs Agents Platform, LiveKit Agents, Retell AI and Vapi state HIPAA compliance or a signed BAA, and the plan matters: in several cases the BAA or HIPAA add-on is on a specific or paid tier, so read the compliance note in each card. For Pipecat, PolyAI, Synthflow, Twilio ConversationRelay and Voiceflow we found a mention of HIPAA-related terms without a BAA, or nothing, so it is shown as "Not stated". Ask for the BAA in writing before any patient data is sent.

Which voice AI platforms keep data in the EU?

Bland AI, Deepgram Voice Agent API, ElevenLabs Agents Platform, PolyAI and Vapi document an EU region or EU endpoint for the product, and in some cases it needs a higher plan or a sales conversation. Retell AI state in their docs that they do not operate services in the EU. For the other 5 tools in the table, EU hosting is not stated, which is different from "not available". Even with an EU region, the speech, language and voice models you connect may run elsewhere, so ask where each step runs. Bland AI, Deepgram Voice Agent API, ElevenLabs Agents Platform and Vapi document both HIPAA and EU data on the pages we read.

Are there open-source voice agent frameworks?

LiveKit Agents (Apache-2.0) and Pipecat (BSD-2-Clause) are open source by the license in their GitHub repositories, and you can run them yourself. Free software still carries costs: the framework is the smaller part, and you pay the speech, language model and telephony providers you connect. Bland AI offer private or on-premises deployment through an Enterprise plan, which is not the same as an open-source project.

What does a voice AI agent cost per minute?

It depends on the platform, and the figures below are as listed on 2026-11-01. Bland AI: Start $0.14/min with $0 platform fee and no card required; Build $0.12/min plus a $299/month platform fee; Enterprise is custom, contracted to your volume. Deepgram Voice Agent API: pay as you go from $0.050 to $0.163 per minute depending on the configuration (Standard $0.075, Advanced $0.163, Custom with your own LLM and TTS $0.050). One rate covers speech-to-text, a managed language model and text-to-speech. A free $200 credit with no card is listed. Growth is prepaid from $4,000+ per year. Enterprise is custom. ElevenLabs Agents Platform: the Free plan includes 15 call minutes; additional call minutes cost $0.08 per minute on every plan, and burst pricing is $0.160 per minute. Included minutes by plan: Starter 75, Creator 275, Pro 1,238, Scale 3,738, Business 12,375. Enterprise is custom. LiveKit Agents: As listed on 2026-11-01 for LiveKit Cloud: Build is $0/month with no card and 1,000 agent session minutes included; Ship includes 5,000 minutes and Scale 50,000, then $0.01 per minute; Enterprise is custom. Self-hosting the framework is free under the license; model and telephony providers bill separately. PolyAI: the platform page lists no prices and says ongoing use is priced per minute, with a "Talk to our team" form. The separate Dialog-RSN-1 API is listed at "2¢ per min", billed per second of connected time, excluding telephony and orchestration, and sales is needed only above 10 concurrent calls. Retell AI: pay as you go shows $0.07-$0.31 / min for AI voice agents with $10 in free credits; Retell Voice Infra $0.055/minute; telephony (Twilio, US) $0.015/min. Enterprise is custom. Synthflow: no published per-minute rate. The page says every new deployment starts on an Enterprise contract scoped to call volume, concurrency, telephony and integrations, and that there is no free plan or trial. Twilio ConversationRelay: $0.07 / minute, with the note "Voice costs are calculated separately." The page is marked "Pricing current as of October 2026" and says custom or high-volume pricing needs sales. Vapi: Vapi hosting is $0.05/min on the pay-as-you-go package, plus provider costs shown as ranges (transcriber $0.0095-$0.0099/min, language model $0.0077-$0.0452/min, voice $0.0146-$0.0238/min) and telephony charged by the provider (for example Twilio inbound $0.008/min). $5 in free credits is listed. Premier is custom, direct sales only. Pipecat and Voiceflow did not publish a per-minute price on the pages we read. Telephony and model costs are often billed separately, so compare the total at your own call volume. We did not compute or estimate any price.

Should I hire a voice AI agency?

It depends on who will own the agent after launch. The platform is the smaller part of the work. Most effort goes into the call flow, connecting your booking or CRM system, testing real calls, handling consent and recording rules, and monitoring after launch. If you have engineers with time for that, a managed platform or an open-source framework can be built in-house. If you do not, a specialist can scope it with you. Our directory lists agencies that name voice AI work on their own websites.

Get matched with 3 vetted agencies

Describe the project. We read every brief and introduce you to three agencies from this index that fit it, usually within 2 working days. Free for buyers; agencies can't pay to be picked.