API Reference
Gemini Decisions
Native Gemini multimodal classification, selection and scoring with probabilities and confidence. Synchronous JSON only; request body limit 20 MiB.
Preview guide
Paste into an AI assistant to help with integration. Documentation and examples only; no real credentials.
Updated · Reviewed
Endpoint and authentication
POST https://api.tu-zi.com/v1beta/decisions
Transport: json
Use an inference API key. Examples contain placeholders; read credentials from environment variables.
Authorization: Bearer YOUR_API_KEY
Content-Type: application/json
Before calling
- Examples use this site's address. For another gateway, replace only the site address and keep the full endpoint path.
- Use an inference key created on this site's token page. Set the
API_KEYenvironment variable or replaceYOUR_API_KEY, as shown in the selected example.
Constraints
This page describes the base protocol. Use the selected model quick start for supported fields, enums, limits and asynchronous behavior. Replace YOUR_MODEL. Form initial values are not upstream defaults.
Minimal request example
{
"model": "gemini-decision-1",
"input": "The ordered mug arrived broken.",
"questions": {
"damaged": {
"type": "noul",
"instructions": "Did the item arrive damaged?"
}
}
}
Confirming success
Check HTTP status, the business status in the body and the required result fields. Do not rely only on HTTP 200 or the script exit code.
Body parameters
| Field | Type | Required | Description / constraints |
|---|---|---|---|
| model | string | Yes | Model ID gemini-decision-1. Example: gemini-decision-1 |
| input | string | object | array | Yes | Shared text or native content blocks for images, audio, video and PDFs. Example: The ordered mug arrived broken. |
| questions | object | Yes | 1–1000 independent questions, each with type and instructions. Choice uses 2–255 criteria entries; score uses 2–255 ordered rubric levels. Example: {"damaged":{"type":"noul","instructions":"Did the item arrive damaged?"}} |
Response fields
| Field | Type | Description / constraints |
|---|---|---|
| answers | object | Typed answers, probabilities and confidence per question. |
| usage.total_input_tokens | integer | Actual input tokens: shared input once plus question instructions and criteria. |
| usage.total_output_tokens | integer | Always zero. |
Illustrative response; some endpoints show essential fields only. Optional fields may be absent.
{
"answers": {
"damaged": {
"type": "noul",
"noul": 0.99,
"confidence": 0.99
}
},
"usage": {
"total_input_tokens": 94,
"total_output_tokens": 0,
"total_tokens": 94
}
}
Errors and retries
Check HTTP status and protocol-specific errors; do not mix error formats across protocols.
- 401/403: check credentials and permissions.
- 429: classify temporary throttling, exhausted quota, or service capacity failure; only transient failures merit bounded backoff.
- 5xx: inspect the gateway, upstream, and model/channel configuration.
Error bodies may contain error/message/code depending on the protocol.
- Use bounded exponential backoff for reads encountering transient errors; fix parameter, authentication, quota, or configuration failures first.
- Do not blindly retry task creation, generation, or key mutations, to avoid duplicate billing or execution.
兔子API