Voice AI built on our own models.
Lokutor is a Madrid company that designs and trains its own speech models, and builds voice agents, text to speech, speech to text and noise cancellation on top of them.
What Lokutor is.
Lokutor designs and trains its own speech models (text to speech, turn-taking, noise suppression and speech recognition) and runs voice agents on them, in its cloud, on customers’ servers or on devices. The models are small and efficient, small enough to run in real time on an ordinary CPU, which is what lets the same stack run in the cloud, on-premise or inside a product. A voice agent costs about 5.2¢ a minute, language model included, paid as you go.
- Founded byDaniel Varela and Guillermo Martin, in Madrid, Spain.
- What we buildVersa 2.0, Turno, Psst, Conv 2.0, Ito. Our models
- ProductsVoice agents, phone agents, text to speech in nine languages, speech to text, noise cancellation and voice cloning, through a dashboard, REST and WebSocket APIs and SDKs for JavaScript and Python.
- Where it runsOur EU cloud, your own servers (on-premise) or inside a product (on-device).
- Open sourceThe Go orchestrator, Oído (speech recognition for a $5 chip) and Ito (a voice for it) are public on GitHub and Hugging Face.
- Programs and partnersBarcelona Supercomputing Center AI Factory, NVIDIA Inception, Madrid Innovation, and an official Pipecat community integration.
- Contactcontact@lokutor.com, or the enterprise form.
Last updated 10 October 2026.
Common questions.
What is Lokutor?
Lokutor is a Madrid company that designs and trains its own speech models and runs voice agents, text to speech, speech to text and noise cancellation on them, in its cloud, on customers’ servers or on devices.
Who founded Lokutor?
Daniel Varela and Guillermo Martin, in Madrid, Spain.
Does Lokutor build its own AI models?
Yes. It trains Versa (text to speech), Turno (turn-taking), Psst (noise suppression), Ito (a voice for microcontrollers) and is developing Conv 2.0 (speech recognition). Production speech recognition today runs NVIDIA’s open Parakeet model on Lokutor’s own servers, and the language model comes from a provider.
Where is Lokutor based and where is data processed?
Lokutor is based in Madrid. Its cloud runs in the EU: calls are processed in AWS Ireland, and accounts and conversation records are stored in Frankfurt.
Is Lokutor open source?
Partly. The Go orchestrator, the Oído speech recognition engine and the Ito voice are open, with GPLv3 code and commercial licences. The Versa voice model and the cloud platform are not.
How do I contact Lokutor?
Write to contact@lokutor.com, or use the form on lokutor.com/enterprise for more calls at once, on-premise or on-device deployments, custom voices and volume pricing.
Try it, or talk to us.
$5 of credit to start, no card.