AI Agents Hub
Vapi
Developer-first platform for building production AI voice agents on phone calls.
The go-to API for devs who want AI phone agents in production — backed by Amazon Ring as a customer, sub-500ms latency, and 1B+ real calls under its belt.
What it is
Vapi is a developer platform for building, testing, and deploying AI voice agents that can make and receive real phone calls. It abstracts the STT → LLM → TTS pipeline into a single API with sub-500ms end-to-end latency, handles telephony, and ships SDKs for web and mobile. Founded in 2023 by Jordan Dearsley and Nikhil Gupta (ex-Waterloo, ex-YC), pivoted from an AI-therapist app to voice infrastructure in 2024. Enterprise customers include Amazon Ring, Intuit, New York Life, Kavak, Instawork and UnityAI; the self-serve developer side has 750k+ devs who have launched 2.5M+ unique agents. 99.9% uptime SLA on enterprise plans.
Best for
- Inbound and outbound AI phone agents (support, sales, scheduling)
- Automated screening, qualification, and routing of inbound calls
- Voice assistants embedded in apps via Voice SDK
- Replace IVR menus with conversational flows in contact centers
Pros
- API-first, very developer-friendly — REST + SDK + dashboard
- Battle-tested at scale: 1B+ calls, 1–5M/day, <500ms avg latency
- Amazon Ring runs 100% of inbound through Vapi — real enterprise reference
- Model-agnostic — swap STT / LLM / TTS providers without rewriting flow
- Compliance built-in: SOC 2, HIPAA, PCI for regulated industries
Cons
- Pay-per-minute costs scale steeply with call volume — model for unit cost
- Voice quality and latency still bound by the chosen LLM/TTS stack
- Mostly an API — you need engineering capacity to wire up the voice flow yourself
Quick start
- Sign up at https://vapi.ai → create a workspace → grab an API key from Settings.
- In the dashboard, click "Create Assistant" → pick the voice (e.g. ElevenLabs "Rachel") and the LLM (e.g. GPT-4o).
- Type the system prompt, e.g. "You are Alex, an appointment scheduler. Ask for the customer's name and preferred date, then book them."
- Click "Test in Browser" to dry-run the conversation (no telephony cost yet).
- Attach a phone number (Vapi provisions one for ~$2/mo, or import your Twilio number) and ship to production.
Sample input / output
Assistant config:
Voice: ElevenLabs "Rachel" (en-US, warm)
STT: Deepgram Nova-2
LLM: GPT-4o, temperature 0.4
System prompt: "You are Alex from Acme Dental. Greet the caller, confirm if they're an existing patient, ask the reason for calling, and offer the next 2 available slots from our schedule. Never quote prices."
Tools: [list_available_slots, book_appointment]Inbound call routed in 320ms, average call length 1m48s
Caller: "I have a toothache"
Alex: "I'm sorry to hear that. Are you an existing Acme Dental patient?"
Caller: "Yes"
Alex: "Thanks. I have Tuesday 10:15 AM with Dr. Lee or Wednesday 2:00 PM with Dr. Park — which works better?"
Caller chose Tuesday. Slot booked, confirmation SMS sent.
Total voice minutes that month: 4,210. Cost (estimate): $0.07/min × 4,210 = $295.Benchmarks
Pricing
Usage-based, per-minute of call (STT + LLM + TTS + telephony). Reference points: 1B+ calls handled to date; Amazon Ring routes 100% of inbound through Vapi since 2025. Raised $50M Series B in May 2026 at ~$500M valuation (lead: Peak XV; M12, Kleiner Perkins, Bessemer participated; $72M total raised). SOC 2 / HIPAA / PCI compliant for enterprise.