9 Vapi Alternatives for AI Phone Agents in 2026

The best Vapi alternatives split into three groups, and the right one depends on how much of the stack you want to own. For a managed platform with a no-code builder, pick Retell AI. For one per-minute price with the models included, pick Bland. For the best-sounding voices with agents attached, pick ElevenAgents. For an open source Vapi alternative you can host yourself, pick LiveKit Agents or Pipecat. If you'd rather build straight on a carrier, look at Telnyx or Twilio.

Most teams don't leave Vapi because the calls are bad. They leave because the bill has five moving parts, because support and concurrency sit behind paid packages, or because a feature they built on was retired. This guide covers what Vapi's own pricing and docs say today, then compares nine options for developers building AI phone agents.

Quick Comparison

Why Teams Look for a Vapi Alternative

Vapi calls itself the developer platform for building voice AI agents. Each assistant combines a speech-to-text model, a language model and a text-to-speech voice, and you choose the provider for each. That flexibility is the reason people pick it, and also the source of most complaints.

The bill has several parts

Vapi's pricing docs put it plainly: on usage billing you pay a per-minute hosting fee, plus the model, voice, transcriber and telephony costs for each call. The pricing page lists the hosting fee at $0.05 per minute and says models pass through at cost.

So the number you see first isn't the number you pay. The page's own calculator shows the hosting fee as only one line of five. If you bring your own provider keys, Vapi stops billing you for that provider and the provider bills you directly, which means more invoices, not fewer.

Support and capacity come in packages

Vapi sells optional Success Packages called Core, Pro and Premier. According to the docs, a package doesn't change how usage is billed. It adds support, service commitments and higher limits.

Two details matter if you run production traffic:

• Usage-only accounts get community support. The docs list documentation, peer-to-peer Discord help and email with no response or uptime SLA. Response-time SLAs start at Core, and an uptime SLA starts at Pro.
• Concurrency is capped by plan. Usage-only includes 4 concurrent call lines, and extra lines are a paid add-on. Vapi's concurrency docs say that when you hit the limit, outbound calls can't start and inbound calls can't be accepted until a slot opens.

HIPAA compliance is also a paid add-on on the pricing page.

Workflows were retired

Vapi's visual Workflows builder is gone. Its migration guide says Workflows was retired on August 18, 2026, and existing workflows stopped running from August 19. Vapi recommends Assistants or Squads instead, and says the Squads pattern has led to better results for customers. If your team built call flows in that builder, you've already done one rebuild.

Where Vapi is still the right call

Vapi remains a good choice when you want to swap speech, model and voice providers freely without running any infrastructure, and when your team is comfortable in an API. If your volume is steady and one of the Success Packages fits, staying is less work than moving.

Managed Platforms: Closest to Vapi

Retell AI: builder plus API

Best for: teams where both engineers and operations people touch the agent.

Retell AI describes itself as an AI voice agent platform for contact centers. It has a no-code conversation flow builder alongside APIs and SDKs, cold and warm call transfers, and batch calling for outbound campaigns.

Pricing is pay as you go and built from parts: a voice infrastructure fee, a text-to-speech rate, an LLM rate that varies by model, and telephony. That's the same shape as Vapi, but Retell shows the summed per-minute estimate up front. The pricing page says every account includes 20 concurrent calls, with more sold per call per month.

The trade-off: less freedom than Vapi to wire in any provider you like. You pick from Retell's list.

Bland: one rate, models included

Best for: teams that want a predictable per-minute number.

Bland leads its pricing page with a direct answer to the pass-through model: no token charges and no model-provider pass-throughs. The per-minute rate includes the language model, speech-to-text and text-to-speech. Telephony is billed separately, on your own carrier or Bland's. Higher tiers add a monthly platform fee for a lower rate, and Enterprise lists on-prem or VPC deployment.

The trade-off: you don't choose the models behind the call. Plans also cap daily calls and concurrency, so check those limits against your peak.

Synthflow: enterprise only now

Best for: large rollouts with a procurement process.

Older lists call Synthflow the no-code, self-serve option. That's out of date. Synthflow's pricing page says it's sold as an annual enterprise agreement, with no free plan or trial and no rate by the minute. Pricing depends on call volume, telephony, integrations and security needs.

The trade-off: you can't sign up and test it this afternoon. For a comparison that includes visual chat builders too, see our Voiceflow alternatives guide.

ElevenAgents: voice quality first

Best for: teams that picked Vapi mainly to use ElevenLabs voices.

ElevenLabs' agent platform is now called ElevenAgents. You'll still see its earlier name, Conversational AI, in older guides. Its docs describe agents that coordinate speech-to-text, a language model, text-to-speech and a turn-taking model, with a visual workflow builder, a knowledge base, a native Twilio integration and SIP trunking. You can choose a supported model or bring your own.

The ElevenAgents pricing page says plans are billed by call minutes, LLM usage is billed separately on top, and ElevenLabs doesn't charge extra for telephony.

The trade-off: you're tied to ElevenLabs for the voice layer. If voices are the reason you're shopping, our ElevenLabs alternatives guide compares text-to-speech providers on their own.

Open Source Vapi Alternatives

LiveKit Agents: the framework with a cloud behind it

Best for: engineering teams that want to own the agent code and choose where it runs.

LiveKit Agents is a framework for building realtime voice AI agents. The framework is Apache-2.0, and its README notes that the turn detection models use a separate LiveKit Model License. It has plugins for speech, language and realtime models, native MCP support, and a Node.js version alongside Python.

For phone calls, LiveKit telephony connects SIP trunks to LiveKit rooms, and you can buy US numbers from LiveKit or bring a third-party SIP provider. Agents can run on LiveKit Cloud or on your own servers. If you self-host, the docs say the SIP service has to be deployed separately.

The trade-off: it's a framework, not a dashboard. You write and deploy code, and you own scaling if you self-host.

Pipecat: composable pipelines in Python

Best for: Python teams that want to control every stage of the audio pipeline.

Pipecat is an open-source Python framework for real-time voice and multimodal agents, licensed BSD-2-Clause and maintained by Daily and the community. You compose a pipeline from modular parts, and the repo lists telephony serializers for Twilio, Telnyx, Plivo, Vonage and others.

You can run it anywhere. Daily also operates Pipecat Cloud, a managed service that deploys your agent image, scales from zero and bills by usage.

The trade-off: the same as LiveKit. More control means more to build, test and monitor.

Build on a Carrier or a Model API

Telnyx: carrier and AI from one vendor

Best for: teams that want fewer vendors between the caller and the model.

Telnyx is a licensed carrier that says it owns a private global network and co-located GPU clusters. Its Voice AI Agents product includes a no-code AI Assistant Builder, and you can use Telnyx-hosted open-source models or bring your own. The product page advertises a single per-minute rate that covers speech-to-text, text-to-speech and orchestration.

The trade-off: a smaller ecosystem of agent-specific tooling than the dedicated platforms.

Twilio Conversation Relay: if you already run on Twilio

Best for: teams with Twilio numbers, call flows and compliance work already in place.

Twilio Conversation Relay handles the speech layer. Twilio turns the caller's speech into text, sends it to a WebSocket server you host, and speaks your response back in a voice you choose. Your application brings the AI. The docs list Deepgram and Google for speech-to-text, and Google, Amazon and ElevenLabs for voices.

The trade-off: it's a building block. Prompting, tools, memory and evaluation are all yours to build.

OpenAI's realtime APIs: speech-to-speech over SIP

Best for: teams that want one speech-to-speech model instead of a three-model pipeline.

OpenAI's Telephony and SIP guide shows how to point a SIP trunk at its API so the model answers the call directly. Your backend gets a webhook for each incoming call and accepts or rejects it. The guide now leads with OpenAI's newer Live API and still documents the Realtime API alongside it.

The trade-off: you bring the SIP trunk and write the call handling. One vendor's model also means no swapping providers mid-project.

How to Choose a Vapi Alternative

Before you commit, run these checks:

1. Price one real call, end to end. Add platform, speech-to-text, model, voice and telephony for a typical call length. Compare totals, not headline rates.
2. Check concurrency at your peak hour. Every platform caps simultaneous calls somewhere. Find the cap and the cost of raising it.
3. Test on real phone lines. Latency and interruption handling feel different over a mobile network than in a browser demo. We're not quoting latency figures here, because vendors measure them on their own terms.
4. Sort out consent early. Call recording and AI disclosure rules vary by region. Our Voiceflow guide linked above has a short section on it.

If you actually want a chatbot on your website rather than a phone agent, start with our Botpress alternatives guide. And if you're looking for a voice assistant to use yourself, not one to build, see AI voice assistants.

Where Taku Fits

Taku isn't a voice agent platform, so it doesn't replace Vapi or any tool on this list. It won't place or answer phone calls for your business. It helps with a different step: you've seen someone share a working AI app or workflow, and you can't get it running yourself.

That's where Taku comes in. Taku Apps and Sites, built by the best AI minds. Find one, run it, and keep it in your Mac Dock. Taku is in Beta, and the Mac app is available now.

FAQ

What is the best Vapi alternative?

For most teams that want a managed platform, Retell AI. Pick Bland for a single per-minute rate with models included, ElevenAgents if voice quality comes first, and LiveKit Agents or Pipecat if you want open-source code you control.

Is there an open source Vapi alternative?

Yes. LiveKit Agents is Apache-2.0 and Pipecat is BSD-2-Clause. Both are frameworks, so you write the agent in code and either self-host it or use the vendor's managed cloud.

How does Vapi pricing work?

Vapi charges a per-minute hosting fee, then adds the cost of the transcriber, language model, voice and telephony for each call. Optional Success Packages add support SLAs and higher limits without changing how usage is billed.

Who are Vapi's main competitors?

Retell AI, Bland and ElevenAgents are the closest managed platforms. LiveKit Agents and Pipecat compete as open-source frameworks. Telnyx and Twilio compete from the carrier side.

Did Vapi remove Workflows?

Yes. Vapi's migration guide says Workflows was retired on August 18, 2026. Vapi recommends rebuilding call flows as Assistants or Squads.

Is Synthflow still a self-serve Vapi alternative?

No. Synthflow's pricing page says it's sold as an annual enterprise agreement, with no free plan or trial.