Language models

OpenAI-compatible chat completions on open-weight Qwen and DeepSeek models.

POST /v1/chat/completions at https://api.albastov.com/v1 returns a model's reply to a list of messages, in the OpenAI chat completions format, with streaming, tool calls, JSON output and thinking.

Quick example

A system message sets instructions for the whole conversation. The request asks qwen3.8-27b for the language code of the user's message, with thinking off.

curl

curl https://api.albastov.com/v1/chat/completions \
  -H "Authorization: Bearer $ALBASTAR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3.8-27b",
    "messages": [
      {"role": "system", "content": "Reply with one ISO 639-1 language code."},
      {"role": "user", "content": "Wie spät ist es?"}
    ],
    "chat_template_kwargs": {"enable_thinking": false},
    "max_tokens": 10
  }'

Python

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.albastov.com/v1",
    api_key=os.environ["ALBASTAR_API_KEY"],
)
reply = client.chat.completions.create(
    model="qwen3.8-27b",
    messages=[
        {"role": "system",
         "content": "Reply with one ISO 639-1 language code."},
        {"role": "user", "content": "Wie spät ist es?"},
    ],
    extra_body={"chat_template_kwargs": {"enable_thinking": False}},
    max_tokens=10,
)
print(reply.choices[0].message.content)

JavaScript

const res = await fetch("https://api.albastov.com/v1/chat/completions", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.ALBASTAR_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    model: "qwen3.8-27b",
    messages: [
      { role: "system", content: "Reply with one ISO 639-1 language code." },
      { role: "user", content: "Wie spät ist es?" },
    ],
    chat_template_kwargs: { enable_thinking: false },
    max_tokens: 10,
  }),
});
if (!res.ok) throw new Error(`${res.status} ${await res.text()}`);
const reply = await res.json();
console.log(reply.choices[0].message.content);

The response, abridged to the reply (id, created and usage left out):

JSON

{
  "object": "chat.completion",
  "model": "qwen3.8-27b",
  "choices": [
    {
      "index": 0,
      "message": {"role": "assistant", "content": "de", "reasoning": null},
      "finish_reason": "stop"
    }
  ]
}

What you can build

Chat completions

Conversations with a system prompt, history, prompt caching and parallel requests.

Streaming

Replies token by token as server-sent events.

Tools and structured output

Function calls with tools, and JSON output that matches a schema.

Thinking

Reasoning before the answer, on or off per request.

Chat models

Every chat model with its context length, parallel requests, thinking switch and tool support.

Key concepts

OpenAI compatibility: The OpenAI SDKs work with base_url set to https://api.albastov.com/v1 and a key with the scope llm. Embeddings, image input, audio and legacy completions are not supported.
Model: model names one chat model. Without model, the request runs on qwen3.8-27b. GET /v1/models lists every model with its state and context_length.
Model loading: One chat model is loaded at a time across the API. A request for another model waits while that model loads, or returns 503 model_loading with Retry-After.
Thinking: Thinking is on by default; chat_template_kwargs.enable_thinking turns thinking off per request. The reasoning arrives in message.reasoning.
Conversation state: The API keeps none. Every request sends the whole conversation, up to the model's context_length tokens of prompt plus output.

Next steps

Quick start

A key, a first chat completion and its reply.

Authentication

Keys, the scope llm and the Authorization header.

Chat completions reference

Every parameter, header, response field and error of POST /v1/chat/completions.

List models reference

The fields and states of GET /v1/models.