Docs/Coding agents

Kilo Code CLI

Use the Kilo Code CLI with your own Osarai GPU through an OpenAI-compatible provider in kilo.json.

Kilo Code is an open source coding agent for VS Code, JetBrains and the terminal. Its CLI talks to Osarai through an OpenAI-compatible provider over Chat Completions.

#Install

npm install -g @kilocode/cli

The package installs the kilo command (kilocode works too).

#Configure

Create the global configuration file, which applies to every project. In our tests the CLI read providers only from this file, not from a kilo.json in the project.

{
  "$schema": "https://app.kilo.ai/config.json",
  "provider": {
    "osarai": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Osarai",
      "options": {
        "baseURL": "https://osarai.xyz/v1",
        "apiKey": "{env:OSARAI_API_KEY}"
      },
      "models": {
        "qwen3.8-27b": {
          "name": "Qwen3.8 27B",
          "tool_call": true,
          "reasoning": true,
          "limit": { "context": 131072, "output": 16384 }
        }
      }
    }
  },
  "model": "osarai/qwen3.8-27b"
}

• {env:OSARAI_API_KEY} reads the key from your environment, so the file holds no secret.
• model is the provider id, a slash, and the key under models.
• limit.context tells Kilo when to compact; set it to your instance's context window.

#Run

export OSARAI_API_KEY="qck_your_key_here"
cd your-project
kilo

To run one prompt without the interface:

kilo run "Explain what src/math.ts does"

#In the VS Code extension

The Kilo Code extension can use the same instance: open its settings, add an OpenAI Compatible provider, and enter the base URL https://osarai.xyz/v1, your key and the model id qwen3.8-27b. Set the context window to 131072 in the model's advanced settings.

#Troubleshooting