Get an API key

European voice AI. On the line in 39 languages.

Elbic turns what your agent writes into natural speech, fast enough for a live call and careful with the details callers notice: account numbers, booking codes, a plain “yes”. Built, trained and hosted in the EU.

Your agent sendsDE · IBAN

Ihre IBAN lautet DE89 3704 0044 0532 0130 00.

The caller hears
  1. Ihre IBAN lautet
  2. DE89
  3. 3704
  4. 0044
  5. 0532
  6. 0130
  7. 00
Your agent sendsEN · Booking code

Your booking code is <spell>KX7Q2M</spell>.

The caller hears
  1. Your booking code is
  2. K
  3. X
  4. 7
  5. Q
  6. 500 ms pause
  7. 2
  8. M
Your agent sendsFR · Phone number

Rappelez-nous au 01 99 00 42 17.

The caller hears
  1. Rappelez-nous au
  2. 01
  3. 99
  4. 00
  5. 42
  6. 17
Your agent sendsIT · Timed pause

Un attimo, per favore. <break time="500ms"/> Controllo subito.

The caller hears
  1. Un attimo, per favore.
  2. 500 ms pause
  3. Controllo subito.
Your agent sendsES · Order number

Su número de pedido es <spell>A4T9</spell>.

The caller hears
  1. Su número de pedido es
  2. A
  3. 4
  4. T
  5. 9
Your agent sendsPL · IBAN

Numer konta: PL61 1090 1014 0000 0712 1981 2874.

The caller hears
  1. Numer konta:
  2. PL61
  3. 1090
  4. 1014
  5. 0000
  6. 0712
  7. 1981
  8. 2874

First audio: 91 ms

Kugel 3, median. WebSocket streaming, network round trip included, connection setup excluded.

Choose an example language
Get an API keyTalk to sales

Illustration of how a line is grouped when it is spoken. Start free. No credit card.

The details a caller hears first.

A voice agent is judged on small things. Our newest model, Kugel 3.5, was trained on exactly these, and gets them right far more often than the model before it.

Text to Speech

  • Numbers in the groups you wrote.

    An IBAN, a phone number or a customer number is read in the groups you typed, the way a person reads it out on the phone.

    DE89 3704 0044

  • A “yes” that sounds like yes.

    One-word replies come out whole and clear, never clipped, in the same voice every time.

    Yes. Okay. One moment.

  • Codes, one character at a time.

    Put a booking code in a spell tag and every character is spoken on its own, in the order you wrote it.

    <spell>KX7Q2M</spell>

  • Pauses as long as you ask.

    A break tag sets the pause before the next sentence, so a caller has a moment to think or to find a pen.

    <break time="400ms"/>

  • Names said your way.

    Give a name, brand or place its IPA transcription in a pronunciation dictionary, and it is said that way instead of guessed from the spelling.

    Nguyen /ŋwiən/

39 languages from one model.

22 of them European, with native pronunciation. A German voice says English words without slipping into another accent.

  • English
  • Deutsch
  • Français
  • Español
  • Italiano
  • Português
  • Nederlands
  • Polski
  • Čeština
  • Slovenčina
  • Magyar
  • Română
  • Български
  • Українська
  • Svenska
  • Dansk
  • Norsk
  • Suomi
  • Hrvatski
  • Slovenščina
  • Srpski
  • Ελληνικά

Your callers’ data stays in Europe.

Developed, trained and hosted in the EU. For a voice agent that hears names, account numbers and health questions, where the audio goes is not a detail.

Why we build in Europe
Sovereign Cloud
For customers in the EU and EEA, hosting is done by EU companies only, on every plan and without an enterprise contract. No provider with a US parent processes your audio, text or voice references.
No training on your content
What you send is used to answer your request, never to train models.
Zero data retention for content
Text and audio are processed in memory and deleted once the request completes. Voice references you upload for cloning follow their own rules.
On-premise
The models can run in your own data centre, behind your firewall, with the same SDK and API as the hosted service.

Fits the stack you already run.

Official plugins and endpoints for the voice agent frameworks teams use in production.

  • LiveKit AgentsOfficial plugin for Python and TypeScript.
  • PipecatDrop-in TTS service for Python pipelines.
  • VapiBuilt-in custom TTS endpoint, no proxy server.
  • CognigyRegister once as a custom speech vendor in Voice Gateway.
  • Coming from ElevenLabsCompatible API: change the base URL, voice ID and output format.

docs.kugelaudio.com

Make the first call today.

Create an account in a minute and start free, with no credit card. Or talk to us about on-premise hosting, SLAs and committed rates.

Get an API keyTalk to sales