API overview

This page describes the Engy inference API as a whole: its endpoints, authentication, model IDs, and the fields that Engy adds to standard requests and responses. It is intended for developers who integrate Engy into an application or connect an existing OpenAI or Anthropic client.

Base URLCopy Anchor Link

All inference endpoints use the following base URL:

https://api.engy.ai/v1

EndpointsCopy Anchor Link

AuthenticationCopy Anchor Link

Authenticate each request with an API key. API keys start with sk-engy-. To create one, see API keys.

Send the key in the Authorization header:

curl https://api.engy.ai/v1/chat/completions \
  -H "Authorization: Bearer $ENGY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "glm-5.2", "messages": [{"role": "user", "content": "Hello"}]}'

The /v1/messages endpoint also accepts the key in the x-api-key header, as Anthropic clients send it:

curl https://api.engy.ai/v1/messages \
  -H "x-api-key: $ENGY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "glm-5.2", "max_tokens": 1024, "messages": [{"role": "user", "content": "Hello"}]}'

The GET /v1/models endpoint does not require a key. If you send a valid key, the prices in the response reflect any pricing that applies to your account.

Use an existing clientCopy Anchor Link

Engy is compatible with the official OpenAI and Anthropic SDKs. Change the base URL and the API key, and keep the rest of your code.

import os

from openai import OpenAI

client = OpenAI(
    base_url="https://api.engy.ai/v1",
    api_key=os.environ["ENGY_API_KEY"],
)

response = client.chat.completions.create(
    model="glm-5.2",
    messages=[{"role": "user", "content": "Hello"}],
)

print(response.choices[0].message.content)

To send Engy-specific fields, such as the provider object, from the OpenAI Python SDK, pass them in extra_body:

response = client.chat.completions.create(
    model="glm-5.2",
    messages=[{"role": "user", "content": "Hello"}],
    extra_body={"provider": {"zdr": True}},
)

Model IDsCopy Anchor Link

Specify the model in the model parameter. For the list of model IDs, see Models and pricing.

Some clients require model names in vendor/model form. Engy accepts one leading vendor segment and ignores it, so engy/glm-5.2 and glm-5.2 select the same model. The response returns the model name as you sent it.

Output lengthCopy Anchor Link

The max_tokens parameter sets the maximum number of output tokens. On /v1/chat/completions, max_completion_tokens takes precedence over max_tokens if you send both.

Before a request starts, Engy reserves credit for the maximum possible cost, which depends on max_tokens. If you omit max_tokens, the reservation covers the model's full output length. Set a smaller max_tokens to reduce the balance that a request requires. For details, see Credits and payments.

Engy request fieldsCopy Anchor Link

Engy accepts the following fields in addition to the standard request format. They work on /v1/chat/completions, /v1/completions, /v1/messages, and /v1/responses.

Engy removes the provider object before the request reaches the model.

Engy response fieldsCopy Anchor Link

Engy adds the following information to responses:

Non-streaming chat and completion responses contain only the fields id, object, created, model, choices, usage, and x_engy. Other OpenAI fields, such as system_fingerprint, are not returned.

Model-specific behaviorCopy Anchor Link

Some models enforce fixed request contracts. Engy applies them so that results are consistent:

• kimi-k3: Sampling parameters are fixed. Engy always sets temperature to 1.0 and top_p to 0.95. If you send n, presence_penalty, or frequency_penalty, Engy sets them to 1, 0, and 0.
• qwen3.6-35b-a3b and qwen3.8-27b: A system message that is not the first message is sent to the model as a user message.
• All models: If you set tool_choice to none, Engy removes any tool calls from the response. For details, see Tool calling.

Endpoint differencesCopy Anchor Link

The /v1/messages and /v1/responses endpoints translate requests into the chat format. The following parts of these formats are not supported:

The /v1/responses endpoint is stateless. Send the full conversation in input with each request.

Next stepsCopy Anchor Link

• Streaming: Receive responses incrementally.
• Errors and limits: Handle errors and capacity limits.
• Inference API reference: Look up the parameters and response fields of each endpoint.