Using the API & Playground

Want to try it right now?

The API Testing page takes your own key, model and message, shows the same call in curl, Python, Node.js, Go and Java — streaming and non-streaming — and sends the request from the page.

Playground

The playground is for quick online testing of AI models: enter a question or an instruction, chat with the model and inspect the result. Open it from Playground in the left navigation, or at /playground.

1. Choose a group and model: pick an available model group and a specific model below the input box to confirm which model this conversation uses.
2. Set request parameters: click the parameter button to enable and adjust generation parameters such as temperature, top P, frequency penalty, presence penalty and max tokens.
3. Write the prompt: type your question, instruction or prompt in the input box at the bottom.
4. Send the request: once the model, parameters and input look right, click Send.
5. Read the reply: when generation finishes, read the response in the conversation area and keep typing to continue a multi-turn conversation.

API address and authentication

The gateway’s OpenAI-compatible address is usually:

https://your-host/v1

Use an API key as the bearer token:

Authorization: Bearer sk-your-api-key

For chat completions:

curl https://your-host/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-your-api-key" \
  -d '{"model":"gpt-4o-mini","messages":[{"role":"user","content":"Hello"}]}'

SDK samples

Python (OpenAI SDK)

from openai import OpenAI

client = OpenAI(
    api_key="sk-xxxxxxxx",
    base_url="https://your-host/v1",
)

response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)

Native Claude format

curl https://your-host/v1/messages \
  -H "x-api-key: sk-xxxxxxxx" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model": "claude-3-5-sonnet-20241022", "max_tokens": 1024, "messages": [{"role": "user", "content": "Hello"}]}'

Native Gemini format

curl "https://your-host/v1beta/models/gemini-1.5-pro:generateContent?key=sk-xxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{"contents": [{"parts": [{"text": "Hello"}]}]}'

Supported endpoints

Chat and text

Endpoint	Path	Description
Chat completions	POST /v1/chat/completions	OpenAI Chat Completions compatible endpoint. Requires model and messages, and supports streaming or normal responses.
Playground chat	POST /pg/chat/completions	Dedicated endpoint for the web playground. Authenticates with your sign-in session instead of an API key; billing works exactly as it does for normal calls.
Text completions	POST /v1/completions	Legacy text completion endpoint. Requires prompt and model.
Responses API	POST /v1/responses	OpenAI Responses format endpoint, supporting input, tools and response events.
Responses compaction	POST /v1/responses/compact	Compacts the context of a Responses request.
Claude Messages	POST /v1/messages	Anthropic Claude Messages compatible endpoint. Requires model and message content, converted to the upstream protocol where needed.
Codex web search	POST /v1/alpha/search	Standalone Codex web search format, supported by a subset of models only.
Realtime conversation	GET /v1/realtime (WebSocket)	OpenAI Realtime style bidirectional endpoint. The model is usually passed as a query parameter, and the API key can be passed as a bearer token or a WebSocket subprotocol.

Vectors and ranking

Endpoint	Path	Description
Embeddings	POST /v1/embeddings	Converts text or an array of text into vectors. Requires input.
Gemini-compatible embeddings	POST /v1/engines/{model}/embeddings	Legacy Gemini-style embedding path that takes the model name from the URL.
Rerank	POST /v1/rerank	Sorts candidate documents by relevance to a query. The query and the document list must not be empty.
Moderation	POST /v1/moderations	Runs content moderation over text or other input. When no model is given, a default moderation model candidate is used.

Images and audio

Endpoint	Path	Description
Image generation	POST /v1/images/generations	Generates images from a prompt. The available sizes, quality levels and counts depend on the model you pick.
Image editing	POST /v1/images/edits	Edits an image from an input image and a prompt. Supports JSON and multipart/form-data requests.
Legacy edit endpoint	POST /v1/edits	A retained compatibility path that enters the image relay. New clients should prefer /v1/images/edits.
Speech to text	POST /v1/audio/transcriptions	Transcribes audio, forwarded to the matching speech recognition channel in the OpenAI audio format.
Audio translation	POST /v1/audio/translations	Recognizes audio content and returns the translated text.
Text to speech	POST /v1/audio/speech	Converts input text into audio.

Async generation endpoints for video, music and similar are covered under “Async generation and task endpoints” in Logs & Async Tasks.

Model lists

Endpoint	Path	Description
Model list	GET /v1/models	Returns the OpenAI-format model list by default. Anthropic headers return the Claude format, and an x-goog-api-key header or key query parameter returns the Gemini format.
Model details	GET /v1/models/{model}	Returns information about one model, or a model_not_found error object if it does not exist.
Gemini model list	GET /v1beta/models	Returns the available models in the Gemini format, with models and nextPageToken response fields.
OpenAI-format model list	GET /v1beta/openai/models	Returns the model list in the OpenAI format under the Gemini-compatible namespace.

Native Gemini paths

Endpoint	Path	Description
Native Gemini model call	POST /v1beta/models/{model}:{action}	Native Gemini format endpoint supporting generateContent, streamGenerateContent and embedding operations.
Gemini-compatible model call	POST /v1/models/{model}:{action}	Accepts Gemini-style model operations under the /v1 prefix, parsed and converted by the Gemini relay.

Note

Not every model supports every endpoint above. Before calling, check in the Model Marketplace that the model is enabled and available to your group, and look at the capabilities it lists (vision, function calling, streaming, and so on).