Small voice models.Real conversations.
Speech models we train ourselves, small enough to run in our cloud, on your servers or on a device. Voice agents on them, paid by the second.
- 5.2¢
- a minute, all in. The LLM included.
- 596 ms
- to the first word of a reply, median of real calls
Press the mic and talk. Words appear as you speak.
Pick a voice and a language, then just talk. No sign-up.
Your agent, on its own phone number.
No carrier account, no SIP trunk to set up: the number comes from us, inside the dashboard.
-
Buy a number
Local numbers in 7 countries, $5 a month, from the dashboard.
-
Give it to an agent
Pick the agent that answers it. Change it whenever you like.
-
Inbound and outbound
It answers every call, day and night, and calls one number or a whole list.
- Buy a number$5 a month
- Give it to an agentChange it any time
- Inbound and outboundAnswers and calls
Numbers in the US, Spain, the UK, Germany, France, Italy and Portugal. Phone agents
Books into your calendar. Writes to your CRM.
Connect once from the dashboard. The agent finds a free time and sends the invite, knows the caller when the call starts, and logs every call on their record.
-
Google Calendar -
HubSpot -
Salesforce -
Pipedrive -
Zoho CRM -
Attio -
Zoho Bigin -
Capsule -
Intercom -
ActiveCampaign -
Copper -
Fireberry -
Salesflare
Five stages, our own models.
Audio in on the left, a voice out on the right.
- Psst Noise suppression
- Turno Turn-taking
- Conv Speech-to-text
- LLM Your provider
- Versa Speech synthesis
596 ms from the end of your sentence to the first word of the reply, median of 30 production calls on 10 October 2026 · Our models · Where the time goes
Same models. Three places.
Our models are small enough to run in real time on an ordinary CPU, so the same stack runs in our cloud, on your servers or inside your device.
Start on our cloud.
Sign in, pick a voice, ship. Pay as you go, with synthesis, transcription and orchestration in the minute.
Live in minutes · $5 credit, no card · EU-hosted
See pricingInside your perimeter.
The full pipeline on your own servers. Audio never leaves.
ARM64 & x86 · Auditable
Talk to usIn the product itself.
On the compute already inside. No lifetime inference bill.
Works offline · No round-trip
For hardware makersWhat we’re building.
A 4.9 MB text-to-speech model for the ESP32-S3. It streams, runs without a cloud and is verified in emulation. Board timings to come.
English and Spanish speech recognition on a $5 ESP32-S3 at 3.7% word error. Accurate, but about twice slower than real time. Open source.
The plans announced on 23 September 2026 at about 2¢ a minute, replaced on 9 October by a prepaid balance at about 5¢ a minute, model included.
The short answers.
What is Lokutor?
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. A voice agent costs about 5.2¢ a minute, language model included, paid as you go. Built in Madrid. About Lokutor →
How much does a voice AI agent cost?
Published prices run from about 5¢ to about 31¢ a minute depending on what is included. Lokutor’s is about 5.2¢ a minute at a typical prompt, with speech recognition, the voice and the language model included. New accounts get $5 of credit with no card. Voice AI pricing guide →
Is Lokutor cheaper than ElevenLabs?
On published prices, yes: agents about 5.2¢ a minute against ElevenLabs’ 8¢, text to speech $4.10 per million characters against $40 to $80, speech to text $0.084 an hour against $0.22. That is price, not quality, and ElevenLabs covers more languages. ElevenLabs alternative →
Can voice AI run on-premise?
Yes. Lokutor’s models are small enough to run in real time on ordinary servers, so the whole voice agent can run inside your own infrastructure and call audio never has to leave it. On-premise voice AI →
Can voice AI run on a device, offline?
Yes. Lokutor builds with hardware makers toward the whole voice agent running on the device, for robots, vehicles and appliances, and has open models for a $5 ESP32-S3 microcontroller. On-device voice AI →
Which languages does Lokutor speak?
Nine: Spanish, Catalan, Galician, Basque, English, Portuguese, French, Italian and German, in all ten voices or your own cloned voice. Voice agents listen in English, Spanish, Portuguese, French, Italian and German. Text to speech →
Run voice AI where your users are.
$5 of credit to start, no card. Or talk to us about more calls at once, your own servers or your device.
Or book a demo, or write to contact@lokutor.com