BestAIStack
Vapi logo

Vapi

Voice agents for builders — build, deploy, and improve voice AI agents as fast as you ship

Best for

Developer teams embedding voice agents into their own product

Key strength

Developer-first depth: a fully exposed API with thousands of configurations, bring-your-own or self-hosted transcriber/LLM/voice models, and tool calling into your own backend.

Pricing starts at

$0.05/min hosting + usage; Core package $29/mo

Try now →

Affiliate disclosure: Some links below may earn us a commission at no cost to you.

Vapi website screenshot

What is Vapi?


Reviewed Vapi's site, pricing, and platform docs as of 2026-10-27; research-only, no agent deployed. Vapi is an API-first platform for building AI phone agents: bring your own transcriber, LLM, and voice (or use theirs), wire in tool calling against your own APIs, and pay per minute for hosting plus pass-through model costs. Developer depth is real — so is the review record's complaint pattern: small but harsh review bases (G2 3/5 across 3 reviews, Trustpilot 2.5/5 across 13) led by support latency and a HIPAA fee that doubled to $2,000/mo.

Key Features


API-native agent building

Thousands of exposed configurations with SDKs, so calling logic lives in your backend instead of a closed dashboard.

Model flexibility

Plug in your own transcription, LLM, and text-to-speech providers — or self-host them — with pass-through provider pricing and no markup.

Tool calling

Agents fetch data and perform actions by calling your APIs mid-conversation, keeping calls tied to your own data.

Testing, simulation & observability

Simulate calls before they reach customers, A/B test prompts and voices, and monitor effectiveness and compliance on every plan.

Pros & Cons


What works

  • Developer-first depth: a fully exposed API with thousands of configurations, bring-your-own or self-hosted transcriber/LLM/voice models, and tool calling into your own backend.
  • Low latency that reviewers confirm feels natural — the vendor claims <500ms average and 1 billion calls supported; a mid-market G2 reviewer says conversations feel natural rather than delayed.
  • Fast first-agent setup even without coding: two of three G2 reviews (both 4.5/5) credit quick setup and an option-rich dashboard — praise that survives even inside the churn review.
  • Simulation and monitoring included on every plan: test calls before production, A/B experimentation on prompts and voices, plus compliance posture (SOC 2, PCI; HIPAA as a paid add-on).

Watch out for

  • Support complaints cluster on both platforms: a Trustpilot reviewer reports roughly week-long waits on production issues; a G2 reviewer got no reply for three weeks across email and Discord.
  • A G2 reviewer documents the HIPAA add-on doubling from $1,000 to $2,000/mo overnight with no warning — and the current page lists HIPAA at $2,000/mo.
  • True cost stacks per minute: hosting ($0.05) + transcriber + LLM + voice + telephony. Vapi's own calculator puts 1,000 min/mo at $82–129 — and Pro requires a $999/mo minimum.
  • Config state bugs reported on Trustpilot: agent settings that don't persist, voice search described as a mess, and live edits with no save state.

Who is it for?


great

Developer teams embedding voice agents into their own product

API-first design, BYO/self-hosted models, tool calling, and a simulation suite — the exact stack developer-first reviewers praise.

good

Startups automating inbound/outbound calls at moderate volume

Core at $29/mo buys 10 concurrent calls and 5 phone numbers; minutes bill separately, so model the per-minute stack before committing.

limited

Healthcare and other regulated industries on small budgets

HIPAA eligibility exists but costs $2,000/mo as an add-on, and one reviewer's team switched vendors over how that price changed.

limited

Non-technical teams wanting hands-off production voice AI

Production support latency and config-state complaints are the recurring review themes; no-code-leaning competitors (Retell AI, Synthflow) carry far larger positive review bases.


Reviewed Vapi's site, pricing, and platform docs as of 2026-10-27. We haven't deployed an agent — everything below comes from public pages and review platforms.

Vapi is a voice-AI platform aimed squarely at developers: wire up an agent through the API, pick a transcriber (Deepgram), an intelligence model (OpenAI), and a voice (ElevenLabs) — or bring your own and self-host — then call your own APIs from inside the conversation via tool calling. The vendor claims 1 billion calls supported, sub-500ms average latency, and 1M+ developers. The differentiator against no-code voice builders is depth: thousands of exposed configurations, simulation suites, and A/B testing come standard.

The Catch

The sticker says $0 to start (or $29/mo for Core). The meter says otherwise. Every minute bills five ways at once: $0.05 hosting, plus transcriber, plus LLM, plus voice, plus telephony transport if you use Twilio instead of Vapi's free SIP. Vapi's own calculator, not ours, lands 1,000 minutes at $82–129/mo. Scaling to the Pro tier means a $999/mo minimum, and HIPAA eligibility is a $2,000/mo add-on — a price one G2 reviewer says doubled from $1,000 overnight, with three weeks of unanswered emails and Discord posts following. That support pattern is the second half of the catch: production phone lines are the kind of thing you need answered fast, and Trustpilot reviewers describe week-long waits plus config-state bugs — settings that reset each other, edits going live unsaved. The review bases are small (G2 n=3, Trustpilot n=13), so read them as signals, not verdicts.

What it actually costs (1,000 min/mo, site calculator)

ComponentMonthly
Vapi hosting ($0.05/min)$50
Transcriber (Deepgram)$10
Intelligence (OpenAI)$8–45
Voice (ElevenLabs)$15–24
Telephony (Vapi SIP)$0
Estimated total$82–129

Packages don't replace this — Core ($29) adds concurrency and support on top; Pro (10% of hosting fee) has a $999/mo floor.

Where it fits

Teams that can code get the most out of it — embedding agents into their own product, tool-calling into internal APIs, simulating before deploy. If you want managed or no-code, G2's own comparison data points at Retell AI (4.8 across 2,640 reviews) and Synthflow (4.5 across 1,016). Complementary stack pieces we've reviewed: ServerAvatar for the servers your backend runs on, Uploadcare for call-recording file handling, and Raycast as the dev-side accelerator.

Vapi is builder-first. Its support, reviewers keep reporting, is Discord-first.

Vapi FAQ


What does Vapi actually cost per minute?
$0.05/min for Vapi hosting plus pass-through model costs — for example Deepgram transcription at ~$0.0095/min, OpenAI intelligence at $0.0077–0.0452/min, and ElevenLabs voice at $0.0146–0.0238/min — plus optional telephony (Twilio $0.008–0.014/min; Vapi's own SIP is free). Vapi's on-site calculator estimates 1,000 minutes at $82–129/mo with no package.
Is the $29 Core plan the whole cost?
No. Core ($29/mo) covers concurrency (10 calls), 5 phone numbers, 30-day retention, ZDR, and email support with a 2-business-day SLA. Minutes still bill at the per-minute stack above, and the Pro tier requires a $999/mo minimum.
How do Vapi's reviews compare with competitors'?
Vapi's own bases are small and harsh — 3/5 on G2 across 3 reviews and 2.5/5 on Trustpilot across 13, as captured 2026-10-27. Competitors Retell AI and Synthflow show 4.8/5 (2,640 reviews) and 4.5/5 (1,016 reviews) on G2 respectively, so no-code-leaning buyers have better-reviewed alternatives; Vapi's edge is developer depth.

Alternatives to Vapi


AI website translation that runs itself

Freemium · €15/mo (Starter, excl. VAT)

Manage cloud servers without a DevOps hire

Freemium · $12/mo (Newbie, list; $10 first month)

File upload infrastructure you don't have to build

Freemium · $66/mo (Pro, billed yearly; $79/mo monthly)

Related articles

All posts

Related resources