OpenAI-Compatible API

ZBStream exposes an OpenAI-compatible Chat Completions API. Any OpenAI SDK or HTTP client that supports a custom base URL works out of the box — cURL, Python, Node.js, and Go included.

Endpoint & Authentication

All OpenAI-compatible requests go to a single base URL and authenticate with a Bearer API key. The model name must exactly match an id returned by GET /v1/models.

Authorization: Bearer $ZBSTREAM_API_KEY
Content-Type: application/json

Parameter	Value	Description
Base URL	https://api.zbstream.com/v1	Base address for all OpenAI-compatible requests
Auth	Authorization: Bearer <key>	A ZBStream API key created in the console
Models	GET /v1/models	List models available to the current API key

Send a Chat Completion

The same non-streaming chat request in four languages. The response follows the OpenAI Chat Completions structure — read choices[0].message.content directly.

curl https://api.zbstream.com/v1/chat/completions \
  -H "Authorization: Bearer $ZBSTREAM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4-pro",
    "messages": [
      { "role": "user", "content": "你好，请介绍一下 ZBStream API。" }
    ],
    "stream": false
  }'

Streaming Responses

Add "stream": true to the request body to receive an SSE stream. Read data events in order and close the connection when the [DONE] marker arrives.

curl https://api.zbstream.com/v1/chat/completions \
  -H "Authorization: Bearer $ZBSTREAM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4-pro",
    "messages": [{ "role": "user", "content": "请逐步回答" }],
    "stream": true
  }'

List Available Models

Call GET /v1/models to list the models available to your API key. Read this list dynamically instead of hard-coding model ids.

curl https://api.zbstream.com/v1/models \
  -H "Authorization: Bearer $ZBSTREAM_API_KEY"

Compatibility Notes

The official openai-python and openai-node SDKs work by simply setting base_url.
Sampling parameters such as temperature and top_p are forwarded as-is; supported ranges depend on the selected model.
Only the model field is rewritten during model mapping — all other fields are forwarded to the upstream channel unchanged.
Fields not documented here may still be forwarded, but are not part of the compatibility guarantee; capabilities follow the current model and channel configuration.