Best AI Voice Agent Platform for Indie Developers (2026)

Verdict: For a first phone or web voice agent, start with ElevenLabs Agents if you want the lowest bundled cost per minute, or Retell if you want one clear per-minute bill and 20 free concurrent calls. Pick Vapi if you want to mix and match providers, and LiveKit Agents or Pipecat if you want open source code you can self-host. Leave Bland for high-volume outbound and Synthflow for enterprise budgets.

Every price and limit below comes from the vendor's own page, checked on 14 Nov 2026. Opinions are ours and marked "Our read".

TL;DR#

• ElevenLabs Agents: Free plan with 15 minutes, Starter $6/mo (75 minutes), Creator $22/mo (275 minutes), Pro $99/mo (1,238 minutes); extra minutes $0.08; LLM billed separately (ElevenLabs Agents pricing).
• Retell: $0 to start, $10 free credits, $0.055/min voice infra plus TTS and LLM per minute; first 20 concurrent calls free (Retell pricing).
• Vapi: $0.05/min hosting, model costs passed through, $5 free credits, 4 concurrent calls on the $0 tier (Vapi pricing).
• Bland: Start plan $0 platform fee at $0.14/min with 10 concurrent calls and 100 calls per day; Build is $299/mo at $0.12/min (Bland pricing).
• LiveKit: Agents framework is Apache-2.0; LiveKit Cloud Build is $0 with 1,000 agent session minutes (LiveKit pricing, GitHub).
• Pipecat: BSD-2-Clause open source framework; Pipecat Cloud is usage-based (GitHub, Pipecat Cloud docs).
• Synthflow: "no free plan or trial and no rate by the minute"; annual enterprise contracts only (Synthflow pricing).

Disclosure: ZeroRaised publishes this comparison and is a launch site for bootstrapped products. ZeroRaised has no commercial relationship with any vendor named here and earns nothing if you sign up. Facts come from the linked sources as checked on 14 Nov 2026.

What this post covers (and what it skips)#

If you are still deciding how a voice agent works (speech-to-text, turn detection, interruptions, self-host vs hosted), read our guide to building an AI voice agent first. If you only need a component, our text-to-speech API comparison covers the voices, and our Twilio alternatives guide covers phone numbers and carriers.

This post is the platform decision: which hosted or open product you sign up for to ship a phone or web voice agent this month, what it costs at indie volume, and where the traps are (concurrency caps, LLM fees billed separately, HIPAA add-ons).

The platforms at a glance#

Sources: each vendor page linked in the sections below.

ElevenLabs Agents: cheapest bundled minutes#

ElevenLabs sells Agents by call minutes rather than its shared credit pool. On the Agents pricing page:

Extra minutes cost $0.08 on paid plans. Burst pricing lets you exceed your concurrency limit, with "excess calls charged at double the standard rate" ($0.16/min). Two caveats: "LLM usage is billed separately on top, based on the model you choose", and "Telephony is included on every plan" means ElevenLabs adds no telephony fee, but your carrier still bills you.

Our read: the best price-to-quality ratio for an indie developer who wants great voices and does not want to assemble a stack. You give up some control over the pipeline.

Retell: the most predictable per-minute bill#

Retell's pricing page says it "starts at $0, pay only for what you use", with "$10 in free credits" and a published range of $0.07 to $0.31 per minute. You pay by component:

• Voice infra: $0.055/min
• Text-to-speech: $0.015/min for Retell platform voices and several partners; ElevenLabs Flash $0.040/min, up to $0.10/min for the top ElevenLabs models
• LLM: per minute by model, for example $0.008/min for Gemini 3.1 Flash Lite up to $0.064/min for Claude 5.5 Sonnet as listed
• Telephony: $0.015/min through Twilio or Telnyx; no charge for SIP trunking or your own telephony

Retell's own calculator shows a "Cost Per Minute" of $0.11 for a configuration with a $0.04/min LLM, $0.055 voice infra and $0.015 TTS. Calls are "tracked to the nearest second". Concurrency is "Free for first 20 concurrency (active calls)", then $8.00 per concurrent call per month, with optional burst at +$0.10/min.

Our read: if you are reselling voice minutes to clients (say, an answering agent for local businesses), Retell's line-item pricing is the easiest to mark up and explain.

Vapi: most flexible, bill you assemble yourself#

Vapi charges "$0.05/min Vapi hosting" and bills model costs separately at provider cost, so your real rate is $0.05 plus whatever your STT, LLM, TTS and telephony providers charge. The $0 tier gives "$5 free credits" and 4 concurrent calls. Core is $29/mo with 10 concurrent calls; Pro is 10% of the hosting fee with a $999/mo minimum and 30 concurrent calls. Extra concurrency is $10 per line per month. HIPAA, with a standard DPA and BAA, is a $2,000/mo add-on.

Our read: choose Vapi when you have strong opinions about providers, or want to bring your own keys and negotiate rates. The $2,000 HIPAA add-on rules it out for most solo health-tech experiments.

Bland: outbound volume, platform fee to scale#

Bland bundles the LLM, transcription and voices into one talk rate ("No token charges"). The Start plan has a "$0 platform fee. No card required", $0.14/min, 10 concurrent calls and caps of 100 calls per day and 100 per hour. Build is "$299/month platform fee" at $0.12/min with 50 concurrent calls and 2,000 calls per day. Transfers cost $0.05 (Start) or $0.04 (Build) per transfer minute, and "Telephony is billed separately, on your own carrier or Bland's at pass-through cost." The BAA is listed as Enterprise only.

Our read: the all-in rate is simple, but at $0.14/min Start is pricier than ElevenLabs or a lean Retell setup, and the per-day cap matters for outbound campaigns.

LiveKit Agents and Pipecat: open source options#

LiveKit Agents is "A framework for building realtime voice AI agents", licensed Apache-2.0, with turn detection models under a separate LiveKit Model License, and it works with LiveKit's telephony stack for phone calls (GitHub). On LiveKit Cloud:

• Build, $0/mo: 1,000 agent session minutes, 5 concurrent sessions, $2.50 of inference credits, 1 free US local number, 50 inbound minutes
• Ship, from $50/mo: 5,000 agent session minutes then $0.01/min, 20 concurrent sessions, $5 inference credits, 100 inbound minutes then $0.01/min
• Scale, from $500/mo: 50,000 agent session minutes, up to 600 concurrent sessions, HIPAA listed

Pipecat is an open source framework "for voice agents, multimodal apps, and realtime AI", licensed BSD-2-Clause (GitHub). You self-host it or deploy to Pipecat Cloud, which the docs say "uses usage-based pricing" (Pipecat Cloud docs); check the pricing page linked there for current rates.

Our read: open source makes sense when voice is your product rather than a feature, when you need custom turn-taking or tools, or when margins matter at scale. For a weekend prototype, a hosted platform ships faster. Our AI agent framework guide covers the text-agent side if your voice agent needs heavier tool use.

OpenAI Realtime API and Synthflow#

The OpenAI Realtime API is a speech-to-speech model rather than a platform: you still need hosting, telephony and call logic (LiveKit and Pipecat both work with it). It is priced per token. Per OpenAI's pricing docs, gpt-realtime-2.1 audio costs $32.00 per 1M input tokens ($0.40 cached) and $64.00 per 1M output tokens; gpt-realtime-2.1-mini costs $10.00 input, $0.30 cached and $20.00 output. We have not converted these to per-minute figures because OpenAI's pricing page does not state tokens per minute; measure on a test call.

Synthflow has moved upmarket. Its pricing page says "every new deployment starts on an Enterprise contract" and "Synthflow has no free plan or trial and no rate by the minute." Our read: not an indie option in 2026.

Worked example: 1,000 minutes per month#

Illustrative arithmetic only, using the rates above. It ignores free credits and taxes, and assumes 1,000 inbound minutes per month within each plan's concurrency limit.

Our read: the headline rates are not comparable until you add what each excludes. Retell's $125 is close to all-in. Vapi's $50 and LiveKit's $59 will rise once you add three model providers. ElevenLabs' $80 plus a small LLM is likely the lowest total at this volume, but check the LLM fees in its dashboard on a real call. To keep the LLM line down, see how to reduce LLM costs.

Decision table#

Our editorial picks, based on the published prices and features above.

US and UK rules for AI phone calls#

United States. On February 8, 2024 the FCC announced a unanimous Declaratory Ruling that "recognizes calls made with AI-generated voices are 'artificial' under the Telephone Consumer Protection Act (TCPA)" (FCC news release). In practice, an AI voice agent placing outbound calls is a robocall under the TCPA, and the release notes that telemarketers need "prior express written consent from consumers before robocalling them."

United Kingdom. The ICO's PECR guidance says "You must not make an automated marketing call" without specific consent, and "General consent for marketing, or even consent for live calls, is not enough." Automated calls must include your name and a contact address or freephone number (ICO telephone marketing guidance). The guidance does not mention AI specifically; our read is that you should treat an AI agent making marketing calls as an automated call.

Our read: start with inbound (support lines, booking, after-hours answering), where the caller started the conversation. Add outbound only with documented consent. This is not legal advice.

How to choose in an afternoon#

1. Decide phone or web. Web-only agents can skip carriers entirely, which makes LiveKit Cloud Build or ElevenLabs Free the easiest start.
2. Estimate minutes and peak concurrent calls. Concurrency caps (4 on Vapi's $0 tier, 5 on LiveKit Build) bite before minute costs do.
3. Build the same 3-minute test call on two platforms with your real prompt and measure the total bill, including LLM.
4. Price your product with a margin over the measured per-minute cost. Our guide on how to price an AI product covers usage-based plans.
5. Check compliance before outbound or health data, not after.

Frequently asked questions#

What is the cheapest AI voice agent platform for an indie developer?#

On published prices checked 14 Nov 2026, ElevenLabs Agents is the cheapest bundled option at low volume: about $80 for 1,000 minutes on Starter or Creator (included minutes plus $0.08 per extra minute), or $99 on Pro, with LLM usage billed on top. If you want to own the code, LiveKit Cloud's free Build plan includes 1,000 agent session minutes, but you still pay for speech, LLM and telephony providers.

Is Vapi or Retell better for a solo developer?#

Retell is easier to budget because it publishes a per-component, per-minute price list and an all-in calculator (its example configuration is $0.11/min). Vapi charges a $0.05/min hosting fee and passes model costs through at provider cost, which gives you more provider choice but a bill you have to assemble yourself. Retell also includes 20 concurrent calls free; Vapi's $0 tier includes 4.

Can I build a voice agent without a hosted platform?#

Yes. LiveKit Agents (Apache-2.0) and Pipecat (BSD-2-Clause) are open source frameworks you can run yourself, connecting your own speech-to-text, LLM and text-to-speech providers. You trade a platform fee for more engineering and hosting work. Both also offer managed clouds: LiveKit Cloud has a free Build plan, and Pipecat Cloud uses usage-based pricing.

Which voice agent platforms offer HIPAA BAAs?#

Vapi lists HIPAA with a BAA as a $2,000/mo add-on. Bland's FAQ says it is HIPAA compliant with a signed BAA, but its table lists the BAA on Enterprise only. ElevenLabs lists BAAs for HIPAA customers on Enterprise. Retell describes itself as HIPAA-ready in its Enterprise section. LiveKit lists HIPAA on its Scale plan table. Confirm with sales before handling health data.

Is it legal to use an AI voice agent to make outbound calls?#

In the US, the FCC's February 2024 Declaratory Ruling treats AI-generated voices as "artificial" under the TCPA, so robocall consent rules apply, including prior express written consent for telemarketing. In the UK, ICO guidance says you must not make automated marketing calls without specific consent for that type of call. Inbound support agents are a much safer starting point. This is not legal advice.

Launch your voice agent#

ZeroRaised is a launch directory for products built without institutional VC. When your agent is answering real calls, work through our guide to launching a bootstrapped app, then submit your product for free. For more comparisons like this one, join the ZeroRaised newsletter.

Sources#

• ElevenLabs Agents pricing, last checked 14 Nov 2026
• ElevenLabs pricing, last checked 14 Nov 2026
• Retell AI pricing, last checked 14 Nov 2026
• Vapi pricing, last checked 14 Nov 2026
• Bland pricing, last checked 14 Nov 2026
• LiveKit pricing, last checked 14 Nov 2026
• LiveKit Agents on GitHub, last checked 14 Nov 2026
• Pipecat on GitHub, last checked 14 Nov 2026
• Pipecat Cloud documentation, last checked 14 Nov 2026
• Synthflow pricing, last checked 14 Nov 2026
• OpenAI API pricing, last checked 14 Nov 2026
• FCC: FCC Makes AI-Generated Voices in Robocalls Illegal, last checked 14 Nov 2026
• ICO: Telephone marketing under PECR, last checked 14 Nov 2026