The protocol for African language AI.
A tokenizer, a consented data network, payments and an app, built as one stack. Members record, choose who can use each recording and are paid. Models read their languages in fewer tokens.
Built so each layer feeds the next: the tokenizer cuts the premium, the network gathers consented recordings, payments reward every accepted minute, and the app puts the choice in the member's hand.
The tokenizer
32,000 African word pieces on Gemma 3's tokenizer: fewer tokens than 12 other tokenizers on all 12 Tier 1 languages.
The consented network
Members record their own languages and choose, recording by recording, who can use each one.
Fair pay
Paid per accepted minute of your own voice, plus a share of licensing revenue. Rates are set per market at launch.
How the tokenizer, network, pay and app fit
Light and dark modes
Nzinga Electric in both themes: black and white grounds, magenta for display, Volt for action and money.
Token results
Premiums against English on FLORES+ devtest, read from the results files. Nothing on this site is estimated.
Invite-only join flow
Phone number, a six-digit code and an invite code from a member. Codes expire after 10 minutes.
Pay that reads like a ledger
An append-only ledger in whole minor units, so totals never drift. Payouts are confirmed by SMS.
Three product pages
Tokenizer, Network and App, each with live figures, interface cards, feature rows and an FAQ.
Company and partners
About, Partners and a role page with the principles, the sovereignty scorecard and open roles.
Start with your own language, on the phone you have
Read the tokenizer results, then try the app. Every recording you make carries your own choice, start to end.
On your terms
Choose who can use each recording, change it any time, and withdraw it whenever you want. You keep what you were paid.