Managed Agents at Previously $0.06 $0.03/minManaged Agents are Previously $0.06 $0.03/min.

ServiceNow#1 on EVA-Bench

Artificial Analysis#1 on Speech Agent Arena

Multilingual AI phone agent that answers in 44 languages.

Sonic, Cartesia's voice model, speaks 44 languages. One agent can follow a caller between 10 of them with language detection, and any other language gets an agent of its own.

Start building

Trusted by leading teams

ServiceNowArtificial Analysis

Ranked #1

on EVA-Bench and Speech Agent Arena

ServiceNow
Indeed
SpeakX
Supernova
1Point1
ELSA

A multilingual answering service on one number

Each call starts with a language and ends with an answer or a handoff in the same one.

The agent looks up the bill once your system verifies the caller, and explains the charge in the caller's language.

What changes when the caller's language changes

Which languages one agent covers depends on the speech settings. The rules on AI notices and recording depend on where the caller is.

Language detection

An agent speaks the language of its voice. Language detection (Beta, in the speech settings) lets one agent follow a caller between English, Spanish, French, German, Hindi, Russian, Portuguese, Japanese, Italian and Dutch. Turning it on moves the agent from Ink-2 to a different speech recognition model. Any other language needs its own agent.

Managed Agents

European Union

Article 50 of the AI Act, which applies from August 2, 2026, covers telling people they are talking to an AI system unless it is obvious. Nuria's first line names her as an AI assistant, and you can write that line for each language you route to her.

Regulation (EU) 2024/1689, article 50

United States

Federal law needs one party's consent to record, but states such as California need all parties. Say the call is recorded, in each language you support, near the start.

18 U.S.C. 2511(2)(d); California Penal Code 632

Spanish text to speech
French text to speech
Hindi text to speech

What follows is not legal advice. Notice and recording rules differ by country and state, so have counsel review your greeting and recording notice for each place you take calls.

Built on best-in-class speech models

Cartesia handles listening, turn-taking and interruptions. You bring the instructions, the LLM and the tools.

  • $0.03per minute
  • #1on Speech Agent Arena and EVA-Bench
  • 24/7answers every call, at any hour

Natural turn-taking

The agent lets callers finish, handles interruptions and pauses, and never talks over anyone.

Your choice of LLM and tools

Pick the LLM that is best suited for your agent's job, and give it tools to take action.

A voice that fits your brand

Pick from a library of natural, expressive voices so every call sounds like your company.

Check that the agent stays in the caller's language

Cartesia scores each finished call on whether the agent kept to the caller's language. A call marked false is the one to replay.

Launch a multilingual phone agent in three steps

Start building

  1. Clone the template

    Copy this use case into your Cartesia account and edit the instructions to match your policies.

  2. Connect your systems

    Connect the agent to the systems it needs, then give it a Cartesia phone number or bring your own.

  3. Test the agent, then go live

    Place test calls from the browser, read the transcripts, and score calls with your own metrics.

Enterprise-grade security

  • HIPAA compliant badge

    HIPAA compliant

  • SOC 2 Type 2 badge

    SOC 2 Type 2

  • GDPR badge

    GDPR

Multilingual AI phone agent FAQs

What is a multilingual AI phone agent?

An AI agent that answers your phone line in the caller's language. Nuria asks which language the caller wants, keeps to it, and hands off to a person or logs a callback in that language.

How much does a multilingual AI phone agent cost?

$0.03/min, plus telephony and LLM usage. A call in Hindi costs the same as one in English. See pricing.

Which languages does it support?

Sonic speaks 44 languages, and each agent speaks the language of its voice. One agent can follow a caller between 10 of them with language detection: English, Spanish, French, German, Hindi, Russian, Portuguese, Japanese, Italian and Dutch. A caller in any other language needs an agent set up for that language.

Can one agent switch languages mid-call?

Among those 10, if you turn on language detection, a Beta setting in the agent's speech settings. It moves the agent to a different speech recognition model than Ink-2, so run test calls in the language pairs your callers actually use. For one line in English and Spanish, see the bilingual answering service.

How do I set up a language the detection does not cover?

Create a separate agent with a voice in that language, and route callers to it by number or by a menu choice. See the voices for Korean, Arabic and Vietnamese, three languages detection does not cover.

What does the sample line show?

One call in Spanish. The sample line keeps to a single language for the whole call, so it can't show a switch. To test switching, build your own agent with language detection turned on and call it.

How does the agent reach our billing and outage systems?

Through webhooks you point at your own systems, such as the billing lookup and the queue directory in this setup. A developer on your side wires each webhook once. Cartesia has no ready-made connector for a utility's billing system.

What happens when no one on shift speaks the caller's language?

The agent checks the queue for that language. If it is closed, the agent logs a callback in that language and doesn't say a person is available unless your system says so.

Find your voice

Talk to an expert.

Connect with a member of our team and learn how Cartesia can help you build world-class voice experiences.

Contact Sales

Start building.

Access our models via API and bring an agent into production with our SDKs and developer tools.

Try Cartesia

We’re hiring!

Build the future of interactive intelligence

Careers page