Gateway

#Inference is paid in $KYZN

The Gateway sells AI inference for $KYZN. It serves open-weight models only, through hosts whose terms permit resale.

It has two surfaces. Programs call an OpenAI-compatible endpoint. People use the Gateway view in the app to issue a key, read the model list and copy a request.

Every request is metered and priced at the host's dollar rate plus a 10% margin, quoted in $KYZN. Each settlement window, the Gateway sells the $KYZN needed to pay the host bill and burns the rest. Its settlement address is exempt from the trade tax, so the sale pays no 4% and the full margin burns.

No part of the margin goes to the Reserve, to stakers or to the development wallet. It is removed from supply.

#Keys are wallet-signed

A Gateway key is issued against a wallet signature. The signature proves control of the wallet without moving funds or paying gas.

1. 01
   Connect

   Open the Gateway view with MetaMask, Rabby or Phantom on Robinhood Chain, chain ID 4663. See Using the app.
2. 02
   Sign one message

   The wallet signs a message. No transaction is sent and no gas is paid.
3. 03
   Receive the key

   The key is issued on signature and bound to the signing wallet. The base URL is shown with the key.
4. 04
   Pay in $KYZN

   Every request made with the key is paid in $KYZN from the bound wallet.

Rate limits are enforced per wallet. A wallet's limit covers every request made with its key.

#Requests follow the OpenAI API

The endpoint accepts the OpenAI chat completions request at /v1/chat/completions. Any client that speaks the OpenAI API works after two changes: the base URL and the key.

The base URL is shown with your key. The samples read it from KYZN_GATEWAY_URL and read the key from KYZN_KEY.

The key travels as a bearer token in the Authorization header. The model field takes one identifier from the model table on this page.

curl $KYZN_GATEWAY_URL/v1/chat/completions \
  -H "Authorization: Bearer $KYZN_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4",
    "messages": [{ "role": "user", "content": "Which supply-chain stage binds?" }]
  }'

The JavaScript and Python samples use the OpenAI SDKs with baseURL or base_url set to the Gateway. The model's answer is read from choices[0].message.content, as with any OpenAI-compatible endpoint.

Metering

Each request is metered on its token count. It is billed at the model's price per 1M tokens, set by the pricing rule on this page.

#Only open-weight models are served

Model names appear exactly as published: DeepSeek V4, Kimi K3, GLM-5.3 and Qwen3.8. The identifier is the value of the model field.

Every model is served through a host whose terms permit resale: DeepInfra, Novita or an equivalent host.

Closed models

Closed models are not resold, because their providers' terms do not permit resale. A closed model enters the Gateway only through bring-your-own-key, where the request runs on the caller's own provider key.

#Capacity is prepaid inference

The Gateway's capacity is inference prepaid with third-party hosts.

A host qualifies only when its terms permit resale. DeepInfra and Novita meet this rule, and an equivalent host can replace either.

A window's host bill is the host's dollar rate applied to the tokens served in that window. The host bill is paid in dollars from the $KYZN sold at settlement.

The Gateway view shows Models online, the count of models serving requests, next to Requests served and Burned by usage.

#Prices add a 10% margin

The price of a model is the host's dollar rate for that model plus 10%. The Gateway quotes the price in $KYZN.

price_usd  = host_rate × 1.10
price_kyzn = price_usd ÷ p
margin     = host_rate × 0.10 ÷ p

The margin is one eleventh of the price. At an unchanged $KYZN price, 90.9% of what a caller pays covers the host and 9.1% is burned.

Display rules

• Model names appear exactly as published.
• Each model's price per 1M tokens appears in $KYZN and in dollars, side by side.
• The burn is shown per request.
• Numerals are set in monospace. Percentages carry one decimal and dollar values carry two.

With the inputs above, the Gateway view lists the model at 44,000 $KYZN and $0.88 per 1M tokens, and shows 1,000 $KYZN as the burn for the request.

#The margin is burned

Requests settle in windows. Each window closes with a sale and a burn.

1. 01
   Meter

   Every request in the window is metered and paid in $KYZN at its quoted price.
2. 02
   Sell

   The Gateway sells the $KYZN needed to pay the window's host bill in dollars.
3. 03
   Burn

   A burn transaction destroys the rest. Total supply falls by the burned amount.

sold   = host_bill ÷ p_settle
burned = received − sold

The burn equals the quoted margin when the $KYZN price at settlement equals the price used for the quotes. A lower price at settlement means more $KYZN is sold and less is burned. A higher price means less is sold and more is burned.

#Burns reduce total supply

$KYZN has a fixed supply of 1,000,000,000 and no mint function. A burned token leaves total supply permanently. See Token and launch.

Backing per $KYZN is Reserve value divided by total supply. A Gateway burn lowers total supply and leaves Reserve value unchanged, so backing per $KYZN rises.

backing before = R ÷ S
backing after  = R ÷ ( S − b )

Redemption pays a share measured against total supply. After a burn, each remaining $KYZN redeems for a larger share of every Reserve asset. See Redemption and The Reserve.

The Burned by usage counter totals every Gateway burn. It appears on the dashboard and in the app's Overview and Gateway views.

#Parameter sheet