Generate replies from a messages array, using either a regular JSON response or an SSE stream.
Request
| Item | Description |
|---|---|
| Method | POST |
| Path | /v1/chat/completions |
| Gateway URL | https://ailink.nativex.com |
| Authentication | Authorization: Bearer sk-xxxx; replace the placeholder with your API Key |
| Request format | application/json |
Replace MODEL_ID with a model from the Model Marketplace that supports this endpoint. Available models, parameters, and permissions depend on your account.
Parameters
| Parameter | Location | Type | Required | Description |
|---|---|---|---|---|
model |
Body | string | Yes | A model that supports chat |
messages |
Body | object[] | Yes | Message array. Common roles are system, user, and assistant. |
stream |
Body | boolean | No | true enables SSE. Default: false. |
temperature |
Body | number | No | The traditional OpenAI range is 0–2. Some reasoning models do not support customization. |
top_p |
Body | number | No | Nucleus sampling parameter; traditional range: 0–1 |
max_completion_tokens |
Body | integer | No | Output token limit; a positive integer supported by the model |
max_tokens |
Body | integer | No | Legacy output limit. Do not set it together with max_completion_tokens. |
tools |
Body | object[] | No | Tool definitions; requires model support |
tool_choice |
Body | string / object | No | Tool selection policy |
response_format |
Body | object | No | Output structure, such as JSON; requires model support |
For multi-turn conversations, include the message history in order. Parameter ranges are model-specific; protocol compatibility does not imply support for every parameter.
Example
curl -X POST 'https://ailink.nativex.com/v1/chat/completions' \
-H 'Authorization: Bearer sk-xxxx' \
-H 'Content-Type: application/json' \
-d '{
"model": "MODEL_ID",
"messages": [
{
"role": "user",
"content": "Introduce yourself in one sentence."
}
],
"stream": false
}'
Response
The following is a simplified response for readability. IDs, URLs, text, and usage values are illustrative.
{
"id": "chatcmpl_example",
"object": "chat.completion",
"choices": [
{
"index": 0,
"message": {
"role": "assistant",
"content": "Hello! How can I help you?"
},
"finish_reason": "stop"
}
]
}
| Field or result | Description |
|---|---|
| choices[].message | Complete reply and any tool calls |
| choices[].finish_reason | Reason generation stopped |
| usage | Usage information returned by the model |
Error handling
For unsuccessful responses, check the HTTP status and error body first. Verify the API Key, model name, account balance, and parameters. If rate limited, reduce concurrency and retry later.
Streaming
Set stream to true and add -N to curl. Read choices[].delta from each data: event instead of parsing the entire stream as one JSON object. A common end marker is data: [DONE].