Your Qwen.
One API.

Capabilities

This deployment serves qwen3.8-27B-14GB-IQ4:latest (Qwen3.5 architecture, ~27B, IQ4_XS) on a private GPU. Ollama reports native support for completion, thinking, and tools.

Quickstart

1. Set the base URL

   Use https://api.qwenlab.ir/v1 (include /v1).

2. Authenticate

   Header Authorization: Bearer YOUR_API_KEY or X-API-Key: YOUR_API_KEY. Never publish real keys.

3. Pick the model

   qwen3.8-27B-14GB-IQ4:latest

4. Connect your AI coding agent

   Select OpenAI Compatible in Kilo Code, Cline, or your SDK and point it to /v1.

BASE_URL=https://api.qwenlab.ir/v1
API_KEY=YOUR_API_KEY
MODEL=qwen3.8-27B-14GB-IQ4:latest

Public endpoints

Only these paths are reverse-proxied to the inference gateway. All other paths serve static docs or return 404.

Text chat

Standard OpenAI chat format: an array of messages with roles system, developer, user, assistant, and tool. Both string and array content parts are supported.

{
  "model": "qwen3.8-27B-14GB-IQ4:latest",
  "messages": [
    { "role": "system", "content": "You are a precise coding assistant." },
    { "role": "user", "content": "Write a typed Python palindrome helper." }
  ],
  "temperature": 0.2,
  "max_tokens": 400,
  "stream": false
}

Streaming

Set "stream": true. The response is text/event-stream with OpenAI-style data: chunks, optional final usage, and [DONE].

curl -N https://api.qwenlab.ir/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3.8-27B-14GB-IQ4:latest",
    "messages": [{"role":"user","content":"Outline a REST error taxonomy."}],
    "stream": true,
    "stream_options": {"include_usage": true},
    "max_tokens": 250
  }'

Thinking / reasoning

This model supports a thinking mode. Turn it on or off per request with OpenAI-compatible fields. Server default is off (OPENAI_THINK=false) for fast coding turns.

• think

  true / false — direct switch (recommended).

• enable_thinking

  Alias of think.

• reasoning_effort

  none → off. minimal | low | medium | high | xhigh → on.

• chat_template_kwargs.enable_thinking

  Nested bool sent by some Qwen clients.

When thinking is on, responses and stream deltas include reasoning_content and reasoning alongside content.

{
  "model": "qwen3.8-27B-14GB-IQ4:latest",
  "messages": [{ "role": "user", "content": "Say OK" }],
  "think": false,
  "max_tokens": 256
}

{
  "model": "qwen3.8-27B-14GB-IQ4:latest",
  "messages": [{ "role": "user", "content": "Plan a safe DB migration." }],
  "reasoning_effort": "high",
  "max_tokens": 2048
}

Tool calling

Pass OpenAI-compatible tools and tool_choice. When the model returns tool_calls, execute them in your agent and reply with a tool role message matching tool_call_id.

{
  "model": "qwen3.8-27B-14GB-IQ4:latest",
  "messages": [
    { "role": "user", "content": "Convert 72 F to Celsius using the tool." }
  ],
  "tools": [{
    "type": "function",
    "function": {
      "name": "fahrenheit_to_celsius",
      "description": "Convert °F to °C",
      "parameters": {
        "type": "object",
        "properties": {
          "f": { "type": "number" }
        },
        "required": ["f"]
      }
    }
  }]
}

Useful parameters

• model

  Model ID: qwen3.8-27B-14GB-IQ4:latest.

• messages

  Conversation turns. Supports string or OpenAI content-part arrays.

• stream

  true for SSE; false for a single JSON object.

• stream_options

  Pass {"include_usage": true} to receive token counts in SSE.

• think / reasoning_effort

  false / none by default; enable per request for deep reasoning.

• temperature

  Lower (0–0.3) for coding & tool use; higher for creative drafting.

• max_tokens

  Cap on generated tokens. Defaults to 8192 when omitted.

• response_format

  Supports json_object and json_schema.

• tools & tool_choice

  Function schemas and choice control (auto, none, or specific function).

Examples

Health check

curl -sS https://api.qwenlab.ir/health

List models

curl -sS https://api.qwenlab.ir/v1/models \
  -H "Authorization: Bearer YOUR_API_KEY"

Persian text

curl -sS https://api.qwenlab.ir/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3.8-27B-14GB-IQ4:latest",
    "messages": [{"role":"user","content":"SSD را در سه جمله توضیح بده."}],
    "stream": false,
    "max_tokens": 300
  }'

Python (OpenAI SDK)

from openai import OpenAI

client = OpenAI(
    base_url="https://api.qwenlab.ir/v1",
    api_key="YOUR_API_KEY",
)

resp = client.chat.completions.create(
    model="qwen3.8-27B-14GB-IQ4:latest",
    messages=[{"role": "user", "content": "Summarize TCP handshake."}],
    stream=False,
    extra_body={"think": False},
)
print(resp.choices[0].message.content)

Clients

Limits & operations

This is a personal GPU relay hosted through an encrypted reverse tunnel to Iran edge infrastructure. The origin PC must be online with the gateway and tunnel healthy. Invalid keys return 401.

If the local GPU is momentarily busy with a background job, chat requests wait up to 30 seconds in queue and return 429 with Retry-After if still occupied.

Contact & API access

Need an API key or integration help for Kilo Code, Cline, or your own client? Reach out with your project details and expected usage.