Read the docs.

Authenticate with Authorization: Bearer sk-coret-.... Keys are issued in the console: sign in, name a key, copy it once. POST /chat/completions and GET /models are live. Every response carries an x-request-id header; quote it when reporting an issue.

curl https://inference.coret.ai/v1/chat/completions \
  -H "Authorization: Bearer $CORET_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "coret/glm-5.3-flash-uncensored",
    "messages": [{"role": "user", "content": "Say hello."}]
  }'

from openai import OpenAI

client = OpenAI(
    base_url="https://inference.coret.ai/v1",
    api_key="sk-coret-...",
)

reply = client.chat.completions.create(
    model="coret/glm-5.3-flash-uncensored",
    messages=[{"role": "user", "content": "Say hello."}],
)
print(reply.choices[0].message.content)

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://inference.coret.ai/v1",
  apiKey: process.env.CORET_API_KEY,
});

const reply = await client.chat.completions.create({
  model: "coret/glm-5.3-flash-uncensored",
  messages: [{ role: "user", content: "Say hello." }],
});
console.log(reply.choices[0].message.content);

Responses are standard OpenAI chat-completion objects. usage reports prompt, completion, and cached token counts. Cached input tokens (prompt_tokens_details.cached_tokens) bill at the lower cached rate automatically.

stream = client.chat.completions.create(
    model="coret/glm-5.3-flash-uncensored",
    messages=[{"role": "user", "content": "Write a haiku."}],
    stream=True,
)
for chunk in stream:
    if chunk.choices and chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="", flush=True)

Set stream: true for token-by-token delivery. The final chunk before [DONE] has an empty choices array and carries usage; guard for it as above.

tools = [{
    "type": "function",
    "function": {
        "name": "get_weather",
        "description": "Current weather for a city.",
        "parameters": {
            "type": "object",
            "properties": {"city": {"type": "string"}},
            "required": ["city"],
        },
    },
}]

reply = client.chat.completions.create(
    model="coret/glm-5.3-flash-uncensored",
    messages=[{"role": "user", "content": "Weather in Lisbon?"}],
    tools=tools,
)
call = reply.choices[0].message.tool_calls[0]
# call.function.name == "get_weather"
# call.function.arguments == '{"city": "Lisbon"}'

Pass OpenAI-shape tools; steer with tool_choice and parallel_tool_calls. Return results as role: "tool" messages to continue the exchange.

reply = client.chat.completions.create(
    model="coret/glm-5.3-flash-uncensored",
    messages=[
        {"role": "system", "content": "Reply with a JSON object."},
        {"role": "user", "content": "Name three primes."},
    ],
    response_format={"type": "json_object"},
)

response_format: {"type": "json_object"} constrains output to valid JSON. Mention JSON in a message as well; models comply more reliably when the prompt asks for it.

Aliases resolve to the same model; use whichever your tooling prefers. GET /v1/models returns the live catalog machine-readably. Prepaid credits, no usage tiers.

Unlisted parameters are ignored, not rejected. Request bodies are capped at 10 MB.

{"error": {"message": "...", "type": "rate_limit_exceeded", "param": null, "code": "rate_limit_exceeded"}}

Accounts run up to 40 concurrent requests by default, with no requests-per-minute cap unless one is set on your account. Over the cap, or during a capacity squeeze, you get 429 with a retry-after header; back off and retry. 5xx responses are retry-safe. Need higher limits? Ask from the console.