API reference

#Base URL

The Gateway API follows OpenAI’s chat completions API. OpenAI’s client libraries work with it once the base URL and the key are set. From a first top-up to a first request, step by step: Use the Gateway.

curl -i https://app.kyzen.network/v1/chat/completions \
  -H "Authorization: Bearer $KYZN_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"flash","messages":[{"role":"user","content":"Which stage binds?"}]}'

What a request costs, and when it is settled: Balance and settlement.

#Endpoints

Every path sits under the base URL. Endpoints marked with a key read the wallet from it; the rest need none.

The three key endpoints, with their requests and answers: Keys.

Chat completions

The answer is OpenAI’s chat completion object, and its model reads the tier’s id. A stream sends data: events and ends with data: [DONE]. When no event has come for 15 seconds, it sends a : keep-alive comment line, which clients skip.

curl -N https://app.kyzen.network/v1/chat/completions \
  -H "Authorization: Bearer $KYZN_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"pro","stream":true,"messages":[{"role":"user","content":"Which stage binds?"}]}'

Account

GET /v1/account answers for the key’s wallet. Amounts are in wei of $KYZN, 18 decimals, as strings; times are milliseconds since 1970, UTC.

Public reads

#Models

The Gateway publishes its models by tier and names no model behind a tier. A request names the tier’s id, and the answer names it back.

GET /v1/models lists each tier in OpenAI’s model shape, with the Gateway’s own fields under kyzen. Each price is the host’s dollar rate plus the 10% margin.

curl https://app.kyzen.network/v1/models

#Errors

Errors take OpenAI’s shape, so OpenAI’s clients raise them as they raise OpenAI’s own. The status code and code name the case; message says what to do.

{
  "error": {
    "message": "This wallet has made 60 requests in the last minute. Wait and retry.",
    "type": "rate_limit_error",
    "code": "rate_limit_exceeded",
    "param": null
  }
}

The type follows the status: authentication_error for 401, permission_error for 403, rate_limit_error for 429, api_error for 500, 502 and 503, and invalid_request_error for the rest. A 429 carries retry-after, in seconds: until the oldest of the 60 chat completions leaves the minute, or 1 while 4 are running.

#Limits

Limits apply per wallet: every key of a wallet shares them. The request limits count chat completions; the other endpoints are not counted.