Integration docs

• Open the console and sign in at https://sub.xiaopangwan.site. The console address and the API Base URL are different addresses with different jobs.
• New here? Register at the same console first, then return to API Keys.
• Open API Keys and create a key. Copy it immediately after creation; the full key is normally shown only once. Save it in a password manager or server-side environment variable. This guide never shows a real key.
• Create separate keys for separate devices or clients. Revoke an old key from the API Keys page and update local environment variables when rotating credentials.
• Check Models & pricing or GET /v1/models for models visible to your account, then replace MODEL_ID in every sample.
• Keep keys in a password manager, operating-system key store, or local environment variable—never in browser code, screenshots, repositories, or chat transcripts.

Raw curl requests use the complete URL. For OpenAI SDKs, set baseURL/base_url to /v1 and do not add another /v1 in method calls.

Shared headers: OpenAI uses Authorization: Bearer your_api_key and Content-Type: application/json; Claude-compatible HTTP uses x-api-key, anthropic-version: 2023-06-01, and Content-Type: application/json; Gemini HTTP uses x-goog-api-key and Content-Type: application/json.

The Claude-compatible HTTP request path is /v1/messages. Do not put /v1/messages into the OpenAI Base URL.

Request GET /v1/models first. After receiving a JSON model list, put one returned ID into the minimal request. Chat Completions returns choices, Responses returns output, Claude Messages returns content, and streaming requests return SSE events. A successful response should contain the result field for its protocol; do not treat HTTP 200 alone as a business-success check.

curl "https://api.xiaopangwan.site/v1/chat/completions" \
  -H "Authorization: Bearer your_api_key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "MODEL_ID",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Open https://sub.xiaopangwan.site, sign in, and create an API key. Copy and save it; local configuration starts with the your_api_key and MODEL_ID placeholders, and never puts a real key in examples, repositories, or chat transcripts.

Request GET /v1/models first, replace MODEL_ID with an ID returned for the current key, then send a minimal request for the selected protocol.

A 2xx response with the selected protocol shape confirms the first request reached the API. This guide does not use a real API key for potentially billable inference requests.

If the first request fails, review Protocols and endpoints, Troubleshooting, and Security and operating guidance in that order.

After signing in, treat the user center as the account source of truth. GET /v1/models returns model IDs visible to the current key.

The public Models & pricing page explains the current display policy. Actual model choice, billing, and balance changes follow the user center and request result.

A model name may be exposed through different protocols or channels. Keep the client protocol, request shape, and MODEL_ID aligned.

The only currently launched image model documented here is gpt-image-2. The openai image-only group charges per image by tier: 1K $0.50, 2K $1.00, and 4K $1.50. Do not attach gpt-image-2 to a regular plus/pro group.

Use the Base URL https://api.xiaopangwan.site/v1 for image calls. The full image-generation URL is https://api.xiaopangwan.site/v1/images/generations, and the full image-edit URL is https://api.xiaopangwan.site/v1/images/edits. The only currently launched model is gpt-image-2.

Only verified parameters are documented: prompt, n, size, and quality. size accepts 1024x1024, 1536x1024, or 1024x1536; quality accepts auto, low, high. Do not guess background/output_format or other unverified value sets.

• Set stream: true for OpenAI Chat Completions and Responses. Consume SSE events incrementally and handle connection loss plus the terminal event.
• Use streamGenerateContent with alt=sse for Gemini. Add stream to the Anthropic Messages request body.
• Image generation uses JSON; image editing uploads multipart/form-data. This guide lists only verified prompt, n, size, and quality fields; it does not infer value sets for background/output_format or other unverified fields.
• Set connection and idle timeouts for streams, and reasonable body limits for image uploads. Do not blindly repeat potentially billable generations after a network error.

curl "https://api.xiaopangwan.site/v1/chat/completions" \
  -H "Authorization: Bearer your_api_key" \
  -H "Content-Type: application/json" \
  -N \
  -d '{
    "model": "MODEL_ID",
    "messages": [{"role": "user", "content": "Hello"}],
    "stream": true
  }'

curl "https://api.xiaopangwan.site/v1beta/models/MODEL_ID:streamGenerateContent?alt=sse" \
  -H "x-goog-api-key: your_api_key" \
  -H "Content-Type: application/json" \
  -N \
  -d '{
    "contents": [{"parts": [{"text": "Hello"}]}]
  }'

curl "https://api.xiaopangwan.site/v1/images/generations" \
  -H "Authorization: Bearer your_api_key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2",
    "prompt": "A calm green landscape",
    "n": 1,
    "size": "1024x1024",
    "quality": "auto"
  }'

curl "https://api.xiaopangwan.site/v1/images/edits" \
  -H "Authorization: Bearer your_api_key" \
  -F "model=gpt-image-2" \
  -F "prompt=Refine the lighting" \
  -F "image=@input.png" \
  -F "n=1" \
  -F "size=1024x1024" \
  -F "quality=auto"

• Why is the URL /v1/v1 or returning 404? Give OpenAI SDKs a Base URL ending at /v1, confirm the client does not append a second /v1, and check /v1 or /v1beta plus endpoint assembly.
• Where does MODEL_ID come from? Call GET /v1/models and replace MODEL_ID with an ID returned for the current key; visibility depends on model, channel, and account permissions. The guide does not fix model names or prices; use the user center and current request result as the source of truth.
• Why do authentication headers differ? OpenAI uses Authorization: Bearer your_api_key; Anthropic uses x-api-key plus anthropic-version; Gemini uses x-goog-api-key.
• Why does a streaming or image request fail? Confirm its request shape matches the model, channel, and permissions. Streaming clients must consume SSE, while image fields and formats follow parameters accepted by the model and channel.
• How should I handle 401, 404, 429, or 5xx? Check authentication, the final URL, rate/balance/quota, or keep a redacted timestamp and status before retrying later; contact service support if 5xx persists.
• What if a key is exposed? Sign in immediately, revoke the old key, create a replacement, and update every client and automation.

• Use environment variables, an operating-system key store, or deployment secrets. Never expose a key in browser code, a repository, screenshot, or chat transcript.
• After a leak, sign in immediately, revoke the old key, create a replacement, and update every client and automation.
• Log only redacted request IDs, timestamps, models, status codes, and latency. Never log authentication-header values.
• Set connection and read timeouts. Use bounded exponential backoff for 429 and recoverable 5xx responses; do not automatically retry authentication or parameter errors.
• Before repeating a generation, edit, or other potentially billable request, determine whether the previous attempt was accepted to avoid duplicate charges.