SeaCloud Model API

Model API provides different invocation protocols for different model types. Multimodal models usually use an asynchronous task protocol: create a task, poll its status, and retrieve the result. Large language models usually use synchronous or streaming protocols, returning text or incremental output within a single request.

LLM

SeaCloud LLM API is for Chat Completions, Anthropic Messages, and Responses API calls. Public endpoints use the unified /llm prefix, and the platform handles API Key authentication, balance checks, and request tracing.

Docs

Base configuration Base URL, authentication, common headers, and SDK configuration.

Chat Completions OpenAI-compatible non-streaming chat completions.

Streaming Responses Chat Completions streaming responses returned as SSE events.

Responses API Explicit OpenAI Responses API public path and response handling.

Errors Common error responses, HTTP status codes, and client handling suggestions.

Quick call

curl

export BASE_URL="https://cloud.seaart.ai/llm"
export API_KEY="YOUR_API_KEY"

curl -sS "${BASE_URL}/v1/chat/completions" \
  -H "Authorization: Bearer ${API_KEY}" \
  -H "Content-Type: application/json" \
  -H "X-Request-Id: demo-$(date +%s)" \
  -d '{
    "model": "gpt-4.1",
    "messages": [
      {
        "role": "user",
        "content": "Introduce SeaCloud LLM API in one sentence"
      }
    ],
    "temperature": 0.7
  }'

Python

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_API_KEY",
    base_url="https://cloud.seaart.ai/llm/v1",
)

response = client.chat.completions.create(
    model="gpt-4.1",
    messages=[
        {
            "role": "user",
            "content": "Introduce SeaCloud LLM API in one sentence",
        }
    ],
)

print(response.choices[0].message.content)

Multimodal models

Multimodal models use an asynchronous queue protocol: create a task, poll its status, and retrieve the result after completion. To stop execution, send a cancellation request. The endpoint in the URL directly uses the SeaCloud model ID, and the request body directly passes the model parameter object.

Lifecycle docs

Base configuration Base URL, authentication, endpoint, and request path.

Create task Create a multimodal task and save the fields needed for later queries.

Poll status Query task status and distinguish waiting, completion, and failure states.

Get result Retrieve image, video, audio, 3D, text, or file outputs.

Cancel task Cancel a task and understand cancellation-related states.

Errors and limitations Common errors, unsupported capabilities, and handling suggestions.

Complete code example

curl

export BASE_URL="https://cloud.seaart.ai/model"
export ENDPOINT="nano_banana_2"
export API_KEY="YOUR_API_KEY"

COMMON_HEADERS=(
  -H "Authorization: Key ${API_KEY}"
  -H "Content-Type: application/json"
)

SUBMIT_RESPONSE="$(curl -sS -X POST "${BASE_URL}/v1/queue/${ENDPOINT}" \
  "${COMMON_HEADERS[@]}" \
  -d '{
    "prompt": "a cinematic photo of a cat astronaut"
  }')"

echo "${SUBMIT_RESPONSE}" | jq .

REQUEST_ID="$(echo "${SUBMIT_RESPONSE}" | jq -r '.request_id')"
STATUS_URL="$(echo "${SUBMIT_RESPONSE}" | jq -r '.status_url')"
RESPONSE_URL="$(echo "${SUBMIT_RESPONSE}" | jq -r '.response_url')"

while true; do
  STATUS_RESPONSE="$(curl -sS "${STATUS_URL}" "${COMMON_HEADERS[@]}")"
  echo "${STATUS_RESPONSE}" | jq .

  STATUS="$(echo "${STATUS_RESPONSE}" | jq -r '.status')"
  ERROR="$(echo "${STATUS_RESPONSE}" | jq -r '.error // empty')"

  if [ "${STATUS}" = "COMPLETED" ]; then
    if [ -n "${ERROR}" ]; then
      echo "task failed: ${ERROR}"
      exit 1
    fi
    break
  fi

  sleep 3
done

curl -sS "${RESPONSE_URL}" "${COMMON_HEADERS[@]}" | jq .

Python

import time
import requests

BASE_URL = "https://cloud.seaart.ai/model"
ENDPOINT = "nano_banana_2"
API_KEY = "YOUR_API_KEY"

headers = {
    "Authorization": f"Key {API_KEY}",
    "Content-Type": "application/json",
}

submit_resp = requests.post(
    f"{BASE_URL}/v1/queue/{ENDPOINT}",
    headers=headers,
    json={
        "prompt": "a cinematic photo of a cat astronaut",
    },
    timeout=30,
)
submit_resp.raise_for_status()
submit_body = submit_resp.json()

request_id = submit_body["request_id"]
status_url = submit_body["status_url"]
response_url = submit_body["response_url"]

while True:
    status_resp = requests.get(status_url, headers=headers, timeout=30)
    status_resp.raise_for_status()
    status_body = status_resp.json()

    status = status_body["status"]
    if status == "COMPLETED":
        if "error" in status_body:
            raise RuntimeError(f"{status_body.get('error_type')}: {status_body['error']}")
        break

    time.sleep(3)

result_resp = requests.get(response_url, headers=headers, timeout=30)
result_resp.raise_for_status()
result_body = result_resp.json()

print("request_id:", request_id)
print("result:", result_body)