Docs/Coding agents

Qwen Code

Run Qwen Code, the coding agent built for Qwen models, on your own Osarai GPU through its OpenAI-compatible auth.

Qwen Code is the Qwen team's coding agent for the terminal. It is tuned for Qwen models, so it pairs well with qwen3.8-27b on your instance, and it talks to Osarai over Chat Completions.

#Install

npm install -g @qwen-code/qwen-code

brew install qwen-code

Qwen Code needs Node.js 22 or newer.

#Configure

Qwen Code uses its OpenAI-compatible auth when three variables are set. Put them in your shell for one session, or in ~/.qwen/.env to keep them:

export OPENAI_API_KEY="$OSARAI_API_KEY"
export OPENAI_BASE_URL="https://osarai.xyz/v1"
export OPENAI_MODEL="qwen3.8-27b"

$env:OPENAI_API_KEY = $env:OSARAI_API_KEY
$env:OPENAI_BASE_URL = "https://osarai.xyz/v1"
$env:OPENAI_MODEL = "qwen3.8-27b"

OPENAI_API_KEY=qck_your_key_here
OPENAI_BASE_URL=https://osarai.xyz/v1
OPENAI_MODEL=qwen3.8-27b

So that the interactive interface skips the sign-in choice and knows the real context window, add ~/.qwen/settings.json:

{
  "security": { "auth": { "selectedType": "openai" } },
  "model": {
    "name": "qwen3.8-27b",
    "generationConfig": { "contextWindowSize": 131072 }
  }
}

A .qwen/settings.json in your project overrides the one in your home directory, but keep the key in ~/.qwen/.env, not in the project.

#Run

cd your-project
qwen

To run one prompt without the interface:

qwen -p "Explain what src/math.ts does"

Inside the interface, /stats shows the tokens used so far and /compress summarises the conversation to free context.

#Troubleshooting