Lokutor

Lokutor

Lokutor is a voice agent platform from Madrid that bundles speech recognition, the language model and the voice into one price of 4 to 5 cents a minute. Its co-founders are Daniel Varela Sánchez and Guillermo Martín-Coello, and its own speech models run on CPUs with no GPUs. Agents run in its EU-hosted cloud, on a company's servers or on-device.

Key Features

• All-in-one minute: Speech-to-text, LLM and voice in one per-minute price, from a free plan with 60 minutes to the $349 Business plan at 4 cents a minute.
• CPU speech stack: Its own models for noise suppression (Psst), turn-taking (Turno), speech-to-text and speech synthesis (Versa); the LLM comes from the customer's provider.
• Latency: The company reports about 139 ms to first audio in streaming mode and 0.66 s to the first reply.
• Languages and voices: 9 languages including Spanish, Catalan, Galician, Basque and Portuguese, with voice cloning from the Growth plan.
• Deployment: Cloud, on-premise on ARM64 or x86 servers, or on-device and offline, plus a Pipecat integration and an open-source orchestrator on GitHub.

Use Cases

• Phone agents: Inbound and outbound call agents with phone numbers at $5 a month and optional CRM lookups.
• Iberian-language apps: Voice products for Spanish, Catalan, Galician and Basque speakers.
• Devices and robotics: Offline voice interaction on the compute already inside a product.

Getting Started

Developers can start on the free plan with no card, or book a demo for on-premise and on-device licensing: https://www.lokutor.com