Our take

The managed, full-pipeline choice — strong when you want call analytics and monitoring out of the box.

What it is

Retell AI is a managed platform for building, testing, deploying, and monitoring production-ready AI voice agents at scale. Like Vapi it provides the full STT→LLM→TTS pipeline, but it leans more into a turnkey managed experience with built-in call analytics, monitoring, and concurrency controls. It is Vapi’s main rival.

Best for

High-volume outbound call campaigns
Contact-center voice automation with monitoring
Voice agents that need analytics and QA

Pros

Managed end-to-end, less to wire up
Strong call analytics and monitoring
Handles concurrency at scale

Cons

Less bare-metal control than a pure API
Cost scales with minutes used

Quick start

Sign up at retellai.com with Google/email → create a workspace → grab a "Live API" or "BYO Twilio" key.
Click "Create Agent" → pick a voice (Cartesia Sonic 2 / ElevenLabs / OpenAI TTS) and LLM (GPT-4o / Claude / Gemini).
Type the system prompt and add tools (functions) — Retell exposes them via function-calling JSON.
Test in the in-browser console (no telephony cost). When happy, connect a phone number (Retell provisions numbers in 50+ countries, or import Twilio).
Roll out: push to a campaign, embed in your IVR via SIP, or trigger from your CRM webhook.
Live monitoring: Retell's dashboard shows concurrency, call success, latency, sentiment per call, full transcript search.

Sample input / output

Input

Retell agent config (sample):

Name: "Acme Claw Appointment Scheduler"
Voice: ElevenLabs "Rachel"
LLM: GPT-4o, temperature 0.4

System prompt:
  "You are Alex from Acme Dental. Greet callers, verify patient identity (DOB + last name),
   check available slots in the Acme scheduling system, and book or reschedule.
   If patient asks about pricing, do NOT quote — offer to transfer to billing.
   Never claim to be human."

Tools (function-calling):
  - get_available_slots(date_window: "YYYY-MM-DD..YYYY-MM-DD")
  - book_slot(slot_id: string, patient_id: string)
  - lookup_patient(dob: "YYYY-MM-DD", last_name: string)
  - transfer_to_billing(reason: string)

Transfer rules:
  - on_keyword(["lawyer", "lawsuit", "manager"]) → warm transfer to human queue

Output

Day-7 production monitoring (10,400 calls):
  Containment (no human): 71%
  Avg call length: 1m 38s
  Avg latency: 412ms (Retell dashboard p95: 720ms)
  Booking success rate: 68% (after identity verification)
  Top failure modes (flagged by Retell's auto-QA):
    - patient_dob miss-typed in 12% of calls → tighter prompt phrased
    - intent "schedule for spouse" not in tool → add get_slots_for_patient_ids
Cost: 10,400 calls × ~$0.085/min average (incl. Twilio number) = $748.84/week

Benchmarks

Pricing $0.07–$0.13 / minute (per-call) — Retell pricing 2026

Concurrency limit Up to 10,000 concurrent calls (Enterprise) — Retell platform

Liveness / latency ~400–700ms typical (p95) — Retell docs

STT options Deepgram, Whisper, Cartesia — Retell product

Customers Roadrunner, Cox Auto, Choice Hotels, mortgage servicing ops — Retell customer pages

BYO telephony Yes — Twilio / Vonage / Telnyx SIP trunks — Retell docs

Pricing

Pay per minute of call

Underlying models

Multiple STT/LLM/TTS providers (model-agnostic)

Should you pick this?

Pick it if You want a managed, monitor-rich voice platform — built-in call analytics, concurrency, QA tools, and white-glove enterprise support out of the box. Strong when your team is operations-led, not engineering-led.

Skip it if You want a low-level API you fully control (Vapi gives you that); or you only need a few voice flows and the cost of concurrency is overkill (single-channel SMB use case).

Retell AI