GatwAI

Retell AI

Explore Retell AI for customer calls and appointments, with component pricing, data handling, limitations and practical pilot workflows.

Provider sources checked

What does Retell AI do?

Retell AI is a platform for building and operating AI phone agents. Businesses can handle inbound conversations, capture appointments, retrieve information and transfer callers to people. Conversation logic, language model, voice and integrations together form the deployed system. This distinguishes the service from a speech generator that simply reads supplied text. The useful comparison concerns the entire call: can the agent understand names, recover from misunderstandings and reach a responsible person when something goes wrong? A pleasant voice does not answer these questions. Retell therefore suits bounded service processes with an accountable operations team, test cases and a verifiable human handoff.
Editorial research · 2026-11-14How we research →

Who is this tool for?

  • Service and operations teams with clearly bounded inbound calls.
  • Developers or integration partners connecting phone agents to existing systems.

Strengths

  • Appointment actions and human handoff are central product capabilities.
  • Call logs and post-call analysis help investigate failures.
  • Configurable models and voices allow different quality and cost profiles.

Limitations and requirements

  • Telephony, destination systems and conversation logic need setup and ongoing ownership.
  • Test interruptions, accents and incorrect information with your own scenarios.
  • Voice quality and vendor benchmarks do not prove error-free handling in your deployment.

Who is behind the product?

Company
Retell AI, Inc.
Headquarters
San Francisco, United States, according to its Y Combinator company profile.
Company founded
2023 according to Y Combinator.
Product launch
February 2024: public launch according to the company's timeline.

The company joined Y Combinator's Winter 2024 batch. Its platform combines conversation orchestration, telephony and external model components.

Suggested ways to use it

Suggested workflows or verification steps. These are not hands-on tests; for unverified products they are not a recommendation to use the service.

Suggested workflow: pilot appointment intake

  1. Define a restricted test calendar, permitted appointment types and a human fallback.
  2. Use test callers to exercise incorrect details, interruptions and appointment changes.
  3. Compare calendar records with call outcomes before admitting real requests.

Goal: A controlled pilot with a verified booking chain; not a GatwAI hands-on test.

Suggested workflow: handle common service questions

  1. Supply approved answers about opening hours, delivery procedures and responsibilities.
  2. Trigger a handoff when information is missing instead of inventing an answer.
  3. Assess failed calls using consistent criteria and retest every change.

Goal: Measurable service quality based on correct answers and successful transfers.

Pricing, allowances and data

What should you check before choosing a plan?

Usage billing with USD 10 initial credit. The advertised USD 0.07–0.31 per voice minute is not a reliable ceiling: infrastructure, voice, model, telephony and extras determine the bill, and some models cost more. Silence is billed and long prompts can add charges. Phone numbers and other options may incur monthly fees; enterprise pricing is quoted.

What should you check about data and permissions?

Under the October 25, 2026 terms, Retell may use de-identified, aggregated call data for model training when recording is not opted out; separate agreements can override this. The privacy notice describes primarily US processing. Call storage is indefinite by default; configure retention, restricted storage mode and protected download links deliberately. Address consent and transfers to connected systems separately.

External reviews and available evidence

Murf employee Soumyadip Banerjee documents his own comparison calls with recordings and analytics on August 25, 2026. He praises Retell's analytics but reports interruption problems and hangups. Murf sells competing voice agents, creating a clear conflict of interest. This is not a GatwAI test.

Reported positives

  • Detailed call analysis in the reported test.

Reported cautions

  • Individual configurations and older models do not establish performance of every current agent.

These reports reflect the linked authors’ experiences. Test date, plan and task affect results; they do not establish a representative overall rating.

Product development and notable changes

  • GPT-Live-1 added to voice agents

    The new voice model works alongside a separate text model for tools. At announcement it supports single-prompt agents on paid accounts, rather than every conversation-flow configuration.

    Source for this change ↗

Which alternative fits which task?

Vapi →

Compare for developer-oriented voice agents with a different component and integration architecture.

Synthflow AI →

Consider as an alternative for visually building business calling workflows.

Frequently asked questions

Do 20 concurrent calls mean 20 free calls?

No. Concurrency describes capacity; call usage remains billable.

Can a Retell agent simply go live on its own?

An agent can technically be deployed. Business approval, consent, failure handling and human availability remain your responsibilities.