Privacy
Last updated 8 August 2026
Geko builds speech models and the API that serves them. This page describes what we collect on this website and through that API, and what we do with it. If anything here is unclear, write to amirlan@geko.sh and we will answer.
This website
geko.sh sets no cookies and runs no analytics, advertising or tracking scripts. Fonts are served from our own domain rather than fetched from a third party, so simply reading these pages sends nothing anywhere except the request for the page itself.
The only thing stored in your browser is a single entry, geko-sound, remembering whether you turned interface sound on or off. It never leaves your device and you can clear it with the rest of your site data at any time.
One exception: the “Talk to us” button opens a scheduling window provided by Cal.com. Loading it contacts Cal.com, and anything you enter to book a call is handled under their privacy policy as well as this one.
Accounts and the API
To give you an account and a working API key, we hold:
- the email address you sign up with, and a hash of each API key rather than the key itself
- one row per request recording the model, the voice, the characters synthesized, the seconds of audio and the time, because that is what your credits are metered against
We do not sell personal data, and we do not share it with anyone except the providers below, or where the law requires it.
We do not keep what you send
The content itself is not stored. Text you send to be spoken is held in memory only for as long as it takes to generate the audio. Audio you send to be transcribed is written to a temporary file on the machine handling that one request, because the decoder reads from a path, and that machine is disposable. Neither is written to our database or to any storage we keep, and neither appears in our logs.
Usage and billing records outlive the request, because they are what your account is metered and invoiced on. They contain counts, never content.
We do not train on your content
Our models are trained on speech we record ourselves, with consent and licensed for training. Nothing you send through the API is used to train or tune them, and the serving systems have no path to write your content anywhere the training pipeline can read it. The only thing that crosses between the two is the finished model, in one direction.
Who else processes it
Running the service takes a short list of providers, and each sees only what it needs to:
- Supabase holds accounts, API key hashes and usage records
- Modal runs the models that produce your result
- Polar is our merchant of record and handles payment, so card details go to them and never through us
- Vercel serves this website and the console
Your rights
Ask us for a copy of what we hold about you, ask us to correct it, or ask us to delete it and close your account. Email amirlan@geko.sh and we will act on it. Deleting your account removes your keys and your account record; billing records are kept where we are required to keep them.
Changes
If this policy changes we will update the date at the top of the page. Material changes to how we handle your data will be sent to the email on your account.
Contact
Geko · amirlan@geko.sh
Need a voice that doesn’t exist yet?
Tell us the language, we record the speakers and train the model