Three peoplein Astana
We founded Geko in July 2026 to build the voice models the large labs have no reason to build: real-time speech for languages with millions of speakers and nothing that works
The company
- Founded
- July 2026
- Where
- Astana, Kazakhstan
- Team
- Three founders, full time
- What we build
- Voice AI for languages global speech models serve badly
- How we build it
- We record and licence the voice data ourselves, then train end to end on it
- Business model
- A self-serve API, metered per character synthesized and per second transcribed
The founders
Who isbehind this
Three founders, working on this full time: the models, the recording pipeline, the API and the console are all ours. Each profile below goes straight to the person.
Where the product standsThe full documentation →
- Stage
- Beta — open self-serve signup, metered billing, $1.00 of credits on every new organisation
- In production
- Since July 2026: Tokay-1.0 for speech, Seta-1.0 for transcription
- Pilots
- Kcell, pleep.app and Oneshott Cafe
- Shipped
- HTTP API, a typed TypeScript SDK on npm, a CLI, and OpenAI-compatible endpoints
- In development
- Chirp, voice agents running on our own two models
- Next
- More languages on the same endpoints
Beta means the API is live, billed and documented, and that we are still changing it in response to the people using it.
Need a voice that doesn’t exist yet?
Tell us the language, we record the speakers and train the model