API Documentation

Endpoint-by-endpoint reference for the X-Serve API.

Base URL

https://api.netpreme.com/v1

The API follows the OpenAI schema, so an OpenAI client works once you change the base URL and the key. Two other dialects are also served, described under their own endpoints below.

Authentication

Every request carries a bearer token:

Authorization: Bearer sk-...

Create keys under Settings > API Keys in the Console. The same key works on every endpoint and every dialect. An x-api-key header isn't accepted, even on the Anthropic-compatible endpoint.

Endpoints

Method	Path	Dialect
POST	/v1/chat/completions	OpenAI Chat Completions
POST	/v1/responses	OpenAI Responses
POST	/v1/messages	Anthropic Messages
POST	/v1/messages/count_tokens	Anthropic Messages
GET	/v1/models	Model discovery
GET	/v1/models/{model}	Model discovery

---

POST /v1/chat/completions

Method	Syntax
POST	https://api.netpreme.com/v1/chat/completions

Description

Generates a chat completion. This is the default endpoint and the one to use unless a client you already have expects one of the other dialects.

Parameters

Name	Type	Req.	Description
model	string	Y	zai-org/GLM-5.2-FP8 or moonshotai/Kimi-K3. Case-sensitive. An unknown value returns 404.
messages	array	Y	Standard OpenAI message objects with role and content.
stream	boolean	N	When true, the reply arrives as server-sent events ending with data: [DONE].
reasoning_effort	string	N	One of none, minimal, low, medium, high, xhigh, max. At none the model answers directly. At every other level it reasons first and returns that separately. An unrecognized value returns 400. Coverage differs by model. zai-org/GLM-5.2-FP8 serves none, high, and max. moonshotai/Kimi-K3 serves none, low, high, and max. The Console draws each model's ladder. Behavior on a level a model doesn't serve differs between the two: zai-org/GLM-5.2-FP8 accepts it and reasons anyway, and moonshotai/Kimi-K3 fails the request. Stay on the levels the model serves.
max_tokens	integer	N	Caps generated tokens. Reasoning counts against this budget.
tools	array	N	Function definitions the model may call. When the model chooses one, finish_reason is tool_calls and message.content is null.

Other parameters in the OpenAI schema pass through to the model.

Request

curl https://api.netpreme.com/v1/chat/completions \
  -H "Authorization: Bearer $NETPREME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "zai-org/GLM-5.2-FP8",
    "reasoning_effort": "none",
    "max_tokens": 40,
    "messages": [{"role": "user", "content": "Say hello in five words."}]
  }'

Response

{
  "id": "99a282d080092df7",
  "object": "chat.completion",
  "created": 1787244180,
  "model": "zai-org/GLM-5.2-FP8",
  "service_tier": null,
  "system_fingerprint": null,
  "choices": [
    {
      "index": 0,
      "finish_reason": "stop",
      "logprobs": null,
      "message": {
        "role": "assistant",
        "content": "Hello, I am saying hello.",
        "reasoning_content": null
      }
    }
  ],
  "usage": {
    "prompt_tokens": 12,
    "completion_tokens": 8,
    "total_tokens": 20,
    "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 0}
  }
}

message.reasoning_content holds the model's reasoning and is null when reasoning_effort is none. usage.prompt_tokens_details.cached_tokens reports how many input tokens came from cache, which matters because X-Serve prices cached input separately.

Tool calling

With tools in the request, the model can answer with a call instead of text. finish_reason is tool_calls and message.content is null:

{
  "finish_reason": "tool_calls",
  "message": {
    "role": "assistant",
    "content": null,
    "tool_calls": [{
      "id": "chatcmpl-tool-a452b272a39007a4",
      "type": "function",
      "function": {"name": "get_weather", "arguments": "{\"city\": \"Paris\"}"}
    }]
  }
}

Run the function yourself, then send the conversation back with a tool message carrying the result. tool_call_id matches the id from the previous response:

"messages": [
  {"role": "user", "content": "What is the weather in Paris?"},
  {"role": "assistant", "tool_calls": [{"id": "chatcmpl-tool-a452b272a39007a4", "type": "function",
    "function": {"name": "get_weather", "arguments": "{\"city\": \"Paris\"}"}}]},
  {"role": "tool", "tool_call_id": "chatcmpl-tool-a452b272a39007a4", "content": "18C, light rain"}
]

The next response carries finish_reason of stop and the answer written from your result. Both models behave the same way. See Inference for the worked exchange.

---

POST /v1/responses

Method	Syntax
POST	https://api.netpreme.com/v1/responses

Description

The OpenAI Responses API. Use it when a client expects that shape. Codex does. The reasoning arrives as an item of type reasoning in the output array rather than as a separate field, and usage appears only in the terminal snapshot.

Request

curl https://api.netpreme.com/v1/responses \
  -H "Authorization: Bearer $NETPREME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "zai-org/GLM-5.2-FP8", "input": "say hi"}'

Response

Abridged, since the object carries the full Responses schema:

{
  "id": "resp_8fa309b927b04b2285f5e8db89720a9f",
  "object": "response",
  "status": "completed",
  "model": "zai-org/GLM-5.2-FP8",
  "output": [
    {"type": "reasoning", "status": "completed", "summary": [{"type": "summary_text", "text": "..."}]},
    {"type": "message", "role": "assistant", "status": "completed",
     "content": [{"type": "output_text", "text": "Hi there! How can I help you today?"}]}
  ],
  "usage": {
    "input_tokens": 14,
    "output_tokens": 121,
    "total_tokens": 135,
    "input_tokens_details": {"cached_tokens": 0},
    "output_tokens_details": {"reasoning_tokens": 0}
  }
}

---

POST /v1/messages

Method	Syntax
POST	https://api.netpreme.com/v1/messages

Description

The Anthropic Messages API. Use it when a client expects that shape. Claude Code does. max_tokens is mandatory here, and reasoning counts against it, so a small value can return a thinking block with no answer.

Parameters

Name	Type	Req.	Description
model	string	Y	zai-org/GLM-5.2-FP8 or moonshotai/Kimi-K3.
max_tokens	integer	Y	Mandatory in this dialect. Allow room for reasoning.
messages	array	Y	Anthropic message objects.

Send anthropic-version: 2023-06-01 alongside the bearer token.

Request

curl https://api.netpreme.com/v1/messages \
  -H "Authorization: Bearer $NETPREME_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "zai-org/GLM-5.2-FP8",
    "max_tokens": 2048,
    "messages": [{"role": "user", "content": "In one sentence, what can you help me with?"}]
  }'

Response

{
  "id": "msg_...",
  "type": "message",
  "role": "assistant",
  "model": "zai-org/GLM-5.2-FP8",
  "content": [
    {"type": "thinking", "thinking": "..."},
    {"type": "text", "text": "..."}
  ],
  "stop_reason": "end_turn",
  "stop_sequence": null,
  "usage": {"input_tokens": 23, "output_tokens": 655}
}

---

POST /v1/messages/count_tokens

Method	Syntax
POST	https://api.netpreme.com/v1/messages/count_tokens

Description

Counts the input tokens a Messages request would consume, without generating anything.

Request

curl https://api.netpreme.com/v1/messages/count_tokens \
  -H "Authorization: Bearer $NETPREME_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "zai-org/GLM-5.2-FP8",
    "messages": [{"role": "user", "content": "Say hello in five words."}]
  }'

Response

{"input_tokens": 9}

---

GET /v1/models

Method	Syntax
GET	https://api.netpreme.com/v1/models

Description

Lists the models your key can call.

Request

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

Response

{
  "object": "list",
  "data": [
<!-- generated:model-list:start -->

    {"id": "zai-org/GLM-5.2-FP8", "object": "model", "created": 0, "owned_by": "Netpreme"},
    {"id": "zai-org/GLM-5.3", "object": "model", "created": 0, "owned_by": "Netpreme"},
    {"id": "moonshotai/Kimi-K3", "object": "model", "created": 0, "owned_by": "Netpreme"}

<!-- generated:model-list:end -->
  ]
}

---

GET /v1/models/{model}

Method	Syntax
GET	https://api.netpreme.com/v1/models/{model}

Description

Returns one model by id.

Request

curl https://api.netpreme.com/v1/models/zai-org/GLM-5.2-FP8 \
  -H "Authorization: Bearer $NETPREME_API_KEY"

Response

{"id": "zai-org/GLM-5.2-FP8", "object": "model", "created": 0, "owned_by": "Netpreme"}

---

Not supported

These parts of the OpenAI schema are reachable but do nothing today. Their errors don't say so plainly, which is why this page names them.

Path	What happens
POST /v1/embeddings	Returns 404 Model not found for either model. X-Serve serves no embedding model.
POST /v1/completions	Returns 200, but the models are chat models and the output isn't usable. Use /v1/chat/completions.
response_format	Accepted on chat completions and ignored. Neither json_schema nor json_object constrains the reply: schema mode returns the same prose as a request without the parameter, and object mode returns JSON wrapped in a Markdown fence. Parse defensively or prompt for the shape you want.

Errors

Errors use the shape of the dialect you called. These are the ones you are most likely to meet.

Status	Example body	Cause
400	{"message":"Failed to deserialize the JSON body into the target type: unknown variant \bogus`, expected one of `none`, `minimal`, `low`, `medium`, `high`, `xhigh`, `max`","type":"Bad Request","code":400}`	A parameter value outside the accepted set. The message names the whole set.
401	{"error":{"message":"Invalid or revoked API Key.","type":"invalid_api_key"}}	The key is wrong, or someone revoked it.
401	{"error":{"message":"Missing API Key. Pass it in the Authorization header as \"Bearer sk-...\".","type":"invalid_api_key"}}	No Authorization header, or an x-api-key header instead of a bearer token.
404	{"error":{"code":404,"message":"The model \zai-org/GLM-5.2-FP8-x` does not exist.","param":"model","type":"NotFoundError"}}| Unknown model id. The id carries no reasoning level, so usereasoning_effort` instead.
409	You can have a maximum of 5 active API Keys. Revoke an active Key before creating another.	You already hold the most active keys allowed.
429	The API Key has exceeded its requests-per-minute limit.	Slow down or ask for a higher limit.

X-Serve doesn't count requests that fail authentication against your usage, because it can't attribute them to an account.

Limits

Limit	Default
Active API keys	5
API key name length	40 characters
Requests per minute	60
Tokens per minute	100,000

Each key carries its own rate limits, so yours may differ from the preceding defaults.

See also

• Quickstart to get a key and send a first request.
• Inference for streaming, reasoning control, and worked examples.
• Available Models for what each model accepts.

To read the list from the API:

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

```json { "object": "list", "data": [

{"id": "zai-org/GLM-5.2-FP8", "object": "model", "created": 0, "owned_by": "Netpreme"},
{"id": "zai-org/GLM-5.3", "object": "model", "created": 0, "owned_by": "Netpreme"},
{"id": "moonshotai/Kimi-K3", "object": "model", "created": 0, "owned_by": "Netpreme"}

] } ```ok