Create decisions

POST /v1/decisions

Implements POST /v1/decisions (the Decisions API).

Body: {"model": "kai", "state": "..."|{...}|[...], "questions": {"<name>": {"type": "choice"|"noul"|"score", "instructions": ..., "criteria": ...}}}. model is kai, Kai's versioned id kai-<12 hex of the weights' sha256> — priced as kai and sent as asked — or Jev by OpenRouter's vendor ids, typesafe/jev-1.13 and ~typesafe/jev-latest, which reach Jev itself and bill at Jev's list price. No Jev id is ever answered by Kai: a bare one, such as jev-latest, is an unknown model. model is required; state and questions are required unless the request names a handle, which carries neither. instructions is optional and any JSON. A choice names at least 2 labels and a score at least 1 level, bounded by the token budget rather than a count; questions holds 1 to 100.

observe holds the state under an id, and a later request naming that id as its handle decides over it again. An id is 1 to 128 characters of A-Z, a-z, 0-9, '.', '_' and '-'. A handle belongs to the org that observed it: no other org's request can name it.

Response: {"id","model","provider","answers":{"<name>":{"type",...}}, "usage":{"input_tokens","output_tokens"},"routing","state_hash","latency_ms"}. usage.input_tokens is the billed count — the request's text counted once, the state once and each question's instructions and options once; a decision over a handle bills the state once, when it was observed.

The body may be sent gzip, deflate, br or zstd encoded; decoded, it is bounded as sent.

Refusals are {"error":{"code","message"}}: 400 malformed JSON or unknown model, 401 no valid credential, 402 insufficient balance, 403 a key kind that may not call this (pk-), 415 any other Content-Encoding, 422 an invalid question or handle id, a state beyond the checkpoint's reach (code state_too_long) or a body past 16 MiB (code request_too_long), 429 rate limited or queue full, 502 the service failed, 503 the model is known and not served, 529 overloaded. 402, 429 and 529 carry Retry-After and Retry-After-Ms, and every answer carries X-Request-Id. Billed on the answer's input tokens at the model's price.

Request

18 fields, body application/json (required).

Response

200 body — 32 fields.

Failure carries the platform error shape — see Errors.

Examples

import { Configuration, AiApi } from 'hanzoai';

const api = new AiApi(new Configuration({ accessToken: process.env.HANZO_API_KEY }));
const { data } = await api.postDecisions({ model: "<model>" });

from hanzoai.cloud import ApiClient, Configuration
from hanzoai.cloud.api import AiApi

client = ApiClient(Configuration(access_token=os.environ["HANZO_API_KEY"]))
result = AiApi(client).post_decisions(model="<model>")

cfg := hanzoai.NewConfiguration()
cfg.AddDefaultHeader("Authorization", "Bearer "+os.Getenv("HANZO_API_KEY"))
client := hanzoai.NewAPIClient(cfg)

resp, _, err := client.AiAPI.PostDecisions(context.Background()).Execute()
if err != nil {
	return err
}

use hanzo_client::apis::{configuration::Configuration, ai_api};

let mut cfg = Configuration::new();
cfg.bearer_access_token = std::env::var("HANZO_API_KEY").ok();

let result = ai_api::post_decisions(&cfg, Default::default()).await?;

import ai.hanzo.cloud.ApiClient;
import ai.hanzo.cloud.api.AiApi;

ApiClient client = new ApiClient();
client.setBearerToken(System.getenv("HANZO_API_KEY"));

var result = new AiApi(client).postDecisions();

curl -X POST https://api.hanzo.ai/v1/decisions \
  -H "Authorization: Bearer $HANZO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
       "model": "<model>"
     }'

AI API · All Hanzo APIs · Interactive reference