LeaoAPI 开发者文档 LeaoAPI Developer Documentation

面向下一代智能化应用的高性能、全托管大模型分布式中转基础设施。

High-performance, fully managed distributed large model routing infrastructure for next-generation AI applications.

LeaoAPI 提供了行业统一的、高可用的算力接入层。我们的网络接口在底层 100% 兼容 OpenAI 官方客户端及 SDK 规范。

LeaoAPI provides an industry-unified, highly available computing access layer. All API endpoints are 100% compatible with OpenAI client specifications.

交互式在线测试沙盒 Interactive API Playground

无需配置开发环境，在下方直接体验 LeaoAPI 极速响应（此处为文档沙盒模拟器）：

Test our API endpoints instantly inside this integrated documentation sandbox simulator:

数据隐私与零缓存承诺 Data Privacy & Zero-Retention Policy

作为面向大厂与高规格出海业务的算力网关，LeaoAPI 严格遵循零知识合规审计：

LeaoAPI adheres to a strict Zero-Knowledge security architecture designed for compliance-sensitive enterprises:

• Payload 100% 透传： 所有流经本网关的提示词及生成内容在底层均采用纯透传协议。系统不设任何持久化日志存储，坚决不缓存用户核心数据。 100% Pass-Through: All payloads route strictly in memory. We run a **Zero-Retention Policy** where no prompts or completions are written to persistent disks.
• 数据训练豁免： 平台已与上游供应商锁定了企业级合规协议，确保您的任何线上商业流数据绝不被用于任何模型的二次训练与微调。 Training Exemption: All connections are wrapped under enterprise commercial clauses, legal binding that guarantees your data is never used for upstream fine-tuning.

快速接入三要素 Three Elements for Quick Access

1. 账户可用余额： 前往 [充值中心] 充值。系统依据实际 Token 消耗精准扣费。 Available Balance: Set up credits in the Dashboard. Charges are calculated dynamically per token.
2. 创建专属令牌 (API Key)： 在 [令牌管理] 申请以 sk-.... 开头的合法身份密钥。 API Key Authentication: Generate your sk-.... token inside the API token panel.
3. 统一终节点 Base URL： https://leaoapi.com/v1 Unified Endpoint Base URL: https://leaoapi.com/v1

认证与鉴权 Authentication and Authorization

在发送 HTTP 请求时，在 Headers 中附加 Authorization 字段：

Attach the standard bearer key inside your request headers:

代码示例：cURL (REST API) Code Example: cURL (REST API)

curl https://leaoapi.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-YOUR_LEAO_API_KEY" \
  -d '{
    "model": "deepseek-r1",
    "messages": [{"role": "user", "content": "Hello."}],
    "stream": true
  }'

代码示例：Python SDK Code Example: Python SDK

from openai import OpenAI

client = OpenAI(api_key="sk-YOUR_API_KEY", base_url="https://leaoapi.com/v1")
response = client.chat.completions.create(
    model="deepseek-v3",
    messages=[{"role": "user", "content": "Hello"}],
    stream=True
)
for chunk in response:
    if chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="")

代码示例：Node.js Code Example: Node.js

import OpenAI from "openai";
const openai = new OpenAI({ apiKey: "sk-YOUR_API_KEY", baseURL: "https://leaoapi.com/v1" });

const stream = await openai.chat.completions.create({
  model: "deepseek-v3",
  messages: [{ role: "user", content: "Hello" }],
  stream: true,
});
for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content || "");
}

代码示例：Go 语言 Code Example: Go Lang

package main

import (
	"context"
	"fmt"
	"github.com/sashabaranov/go-openai"
)

func main() {
	config := openai.DefaultConfig("sk-YOUR_LEAO_API_KEY")
	config.BaseURL = "https://leaoapi.com/v1"
	client := openai.NewClientWithConfig(config)

	resp, _ := client.CreateChatCompletion(
		context.Background(),
		openai.ChatCompletionRequest{
			Model: "deepseek-v3",
			Messages: []openai.ChatCompletionMessage{{Role: "user", Content: "Hello"}},
		},
	)
	fmt.Println(resp.Choices[0].Message.Content)
}

代码示例：Java (OpenAI-Java Client) Code Example: Java (OpenAI-Java Client)

import com.theokanning.openai.service.OpenAiService;
import com.theokanning.openai.completion.chat.ChatCompletionRequest;

public class LeaoClient {
    public static void main(String[] args) {
        // 配置超时与 BaseURLConfigure timeout and BaseURL
        OpenAiService service = new OpenAiService("sk-YOUR_KEY", java.time.Duration.ofSeconds(60));
        // 注：原生框架底层替换 leaoapi.com/v1 即可完美映射Note: Simply replace the base URL with leaoapi.com/v1 in native framework for perfect routing.
        System.out.println("Java SDK Connected successfully to LeaoAPI Layer.");
    }
}

SSE 流式协议与 DeepSeek-R1 思维链响应规范 SSE Protocol & DeepSeek-R1 Reasoning Schema

针对类似 deepseek-r1 的深度推理模型，系统在 SSE 协议块的 delta 结构中增加了标准的 reasoning_content 扩展字段。真实的 Chunk JSON 交互序列如下：

For reasoning models like deepseek-r1, the gateway outputs an explicit reasoning_content parameter inside the delta frame:

// 阶段 1：输出模型深度思考思维链Chunk Phase 1: Output model's reasoning thought process
data: {"id":"ch-1","object":"chat.chunk","choices":[{"delta":{"reasoning_content":"Thinking: The user wants..."},"finish_reason":null}]}

// 阶段 2：思维链结束，开始输出最终文本答案Chunk Phase 2: Core answer conversion after reasoning completes
data: {"id":"ch-1","object":"chat.chunk","choices":[{"delta":{"content":"Here is the direct answer:"},"finish_reason":null}]}

// 阶段 3：网关透传标准终止状态符Chunk Phase 3: Finalizer standard stop flag
data: [DONE]

高级功能：智能体工具回调 (Function Calling) Advanced Feature: Tool & Function Calling

LeaoAPI 完美支持模型动态解析结构化插件并调用外部函数：

LeaoAPI perfectly handles structural JSON tool-call routing schemas out of the box:

# 完美支持 tools 传递参数定义Perfectly supports tools parameter definitions
tools = [{
    "type": "function",
    "function": {
        "name": "get_stock_price",
        "parameters": {
            "type": "object",
            "properties": {"ticker": {"type": "string"}}
        }
    }
}]
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "What's the price of AAPL?"}],
    tools=tools
)
print(response.choices[0].message.tool_calls)

高级功能：严格结构化输出 (JSON Mode) Advanced Feature: Structured Outputs

通过锁定 response_format 约束，保证模型 100% 返回可被程序直接无缝反序列化的 JSON 干净结构化数据体：

Enforce strict data boundaries by utilizing the response_format parameter:

// 传入请求核心控制字段约束API Request body format constraint
{
  "model": "deepseek-v3",
  "messages": [{"role": "user", "content": "List 3 items in JSON format."}],
  "response_format": { "type": "json_object" }
}

图像生成与处理接口 (Images API) Image Generation API

LeaoAPI 底层 100% 兼容 OpenAI 官方图像生成规范。统一接入点支持国内顶尖模型的文生图（Text-to-Image）请求。网关针对高并发图像生成任务进行了分发层异步交付处理，确保 URL 返回的高可用性与强稳定性。

LeaoAPI is 100% compatible with the OpenAI Images specification. Our unified endpoint seamlessly routes high-concurrency Text-to-Image requests to top tier domestic models with advanced asynchronous delivery optimization.

1. 统一图像终节点 Base URL： https://leaoapi.com/v1/images/generations Unified Images Endpoint: https://leaoapi.com/v1/images/generations

代码示例：cURL (REST API)Code Example: cURL

curl https://leaoapi.com/v1/images/generations \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-YOUR_LEAO_API_KEY" \
  -d '{
    "model": "wanx-v2",
    "prompt": "A modern cyberpunk style tech office in Shanghai, realistic cinematic lighting, 4k",
    "n": 1,
    "size": "1024x1024",
    "response_format": "url"
  }'

代码示例：Python SDKCode Example: Python SDK

from openai import OpenAI

client = OpenAI(api_key="sk-YOUR_API_KEY", base_url="https://leaoapi.com/v1")

response = client.images.generate(
    model="hunyuan-dit",
    prompt="A beautiful landscape oil painting of West Lake in Hangzhou, sunset, impressionism style",
    n=1,
    size="1024x1024",
    response_format="url"
)

print(response.data[0].url)

视频生成 / Seedance（火山引擎官方协议） Video Generation / Seedance (Volcengine Native Protocol)

Seedance 视频生成是异步任务：提交请求只返回任务 id，随后查询同一任务直至 succeeded 或 failed。本页保留火山方舟的请求体与路径，仅将官方主机 https://ark.cn-beijing.volces.com 替换为 https://leaoapi.com；请勿把它替换成 /v1/images/generations 或 Chat Completions 端点。

Seedance video generation is an asynchronous task: submission returns an id, then the task is retrieved until it is succeeded or failed. This page preserves Volcengine Ark's request body and path, replacing only https://ark.cn-beijing.volces.com with https://leaoapi.com. Do not use an Images or Chat Completions endpoint.

模型与必要字段Models and required fields

创建任务：cURLCreate task: cURL

curl -X POST "https://leaoapi.com/api/v3/contents/generations/tasks" \
  -H "Authorization: Bearer sk-YOUR_LEAO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "doubao-seedance-2-0-260128",
    "content": [{
      "type": "text",
      "text": "A golden retriever runs along a sunlit beach, cinematic tracking shot, natural motion"
    }],
    "ratio": "16:9",
    "duration": 5
  }'

# Successful submission: {"id":"cgt-..."}

查询任务与取得视频 URL：cURLRetrieve task and video URL: cURL

curl "https://leaoapi.com/api/v3/contents/generations/tasks/cgt-YOUR_TASK_ID" \
  -H "Authorization: Bearer sk-YOUR_LEAO_API_KEY"

# When status is "succeeded", read content.video_url:
# {"id":"cgt-...","status":"succeeded",
#  "content":{"video_url":"https://.../generated.mp4"}}

Python：使用 requests 创建并轮询Python: create and poll with requests

该原生任务协议不是 OpenAI Python SDK 的 Images API；使用 requests 可避免把任务接口误当成同步生成接口。

This native task protocol is not the OpenAI Python SDK Images API. Use requests to avoid treating it as a synchronous generation endpoint.

import os
import time
import requests

BASE_URL = "https://leaoapi.com/api/v3"
headers = {
    "Authorization": f"Bearer {os.environ['LEAO_API_KEY']}",
    "Content-Type": "application/json",
}

payload = {
    "model": "doubao-seedance-2-0-260128",
    "content": [{"type": "text", "text": "A paper boat floats down a quiet stream at dawn"}],
    "ratio": "16:9",
    "duration": 5,
}
create = requests.post(f"{BASE_URL}/contents/generations/tasks", headers=headers, json=payload, timeout=30)
create.raise_for_status()
task_id = create.json()["id"]

deadline = time.monotonic() + 900
while time.monotonic() < deadline:
    task = requests.get(f"{BASE_URL}/contents/generations/tasks/{task_id}", headers=headers, timeout=30)
    task.raise_for_status()
    result = task.json()
    if result["status"] == "succeeded":
        print(result["content"]["video_url"])
        break
    if result["status"] == "failed":
        raise RuntimeError(result.get("error", result))
    time.sleep(5)
else:
    raise TimeoutError(f"Seedance task {task_id} did not finish in time")

规范来源：火山方舟 Seedance 视频生成 API（原生 contents/generations/tasks 协议）。Protocol reference: Volcengine Ark Seedance Video Generation API (native contents/generations/tasks protocol). Volcengine Docs ↗

核心国内图像模型清单标识与单独计费边界 Domestic Image Model Identifiers & Pricing Boundaries

请注意：图像类模型不遵循 Token 计费公式。系统采用行业标准，依据请求生成的张数（N）以及分辨率大小进行独立单张额度扣减，不占用文本类 TPM 限制。

Note: Image models do not follow token billing. Charges are calculated strictly per generated image slice based on resolution, completely separated from text TPM constraints.

企业账户混合计费与财务对账公式 Billing Architecture Formula

为了方便企业内部财务系统以及网关层对账，系统扣费严格遵循公开公式（无额外隐性通道服务费）：

Charges are calculated purely via a transparent formula mapping exact usage dimensions with no hidden cost:

* 提示：对于自带思维链输出的模型，其生成的推理 Token（Reasoning Tokens）全部归入“输出 Tokens 费率”合并计费，不设双重扣费陷阱。Note: Reasoning tokens generated by models like R1 are aggregated cleanly under Output Token rates without surcharges.

核心可用模型清单标识与硬性上游边界 Core Model Identifiers & Context Hard Limits

频率与并发控制分级限制阶梯 Rate Limits Grading Matrix

全局错误码参考 Global Error Codes Reference