Connect Kilo CLI to NixAPI: Global Config and Custom Models

This guide uses custom models in Kilo CLI 1.0 and later, checked against the official CLI documentation on October 30, 2026. It adapts the documented compatible-provider interface to NixAPI; live compatibility still depends on your key and model.

1. Install Kilo

With Node.js and npm available, run:

npm install -g @kilocode/cli

Create a key in the NixAPI console and choose a tool-capable chat model from models and pricing.

2. Set the key

On macOS or Linux:

export NIXAPI_API_KEY='YOUR_NIXAPI_API_KEY'

On Windows PowerShell:

$env:NIXAPI_API_KEY = 'YOUR_NIXAPI_API_KEY'

Replace the placeholder and launch Kilo from that terminal.

3. Use global configuration

Edit ~/.config/kilo/kilo.json, merging with existing settings. Current {env:VAR} interpolation requires trusted/global configuration; an ordinary repository configuration cannot use this secret interpolation in the same way. See the custom model guide.

{
  "$schema": "https://app.kilo.ai/config.json",
  "model": "nixapi/NIXAPI_MODEL_ID",
  "provider": {
    "nixapi": {
      "npm": "@ai-sdk/openai-compatible",
      "options": {
        "baseURL": "https://api.nixapi.com/v1",
        "apiKey": "{env:NIXAPI_API_KEY}"
      },
      "models": {
        "NIXAPI_MODEL_ID": {
          "name": "NixAPI coding model",
          "tool_call": true,
          "limit": { "context": 32768, "output": 4096 }
        }
      }
    }
  }
}

Replace both occurrences of NIXAPI_MODEL_ID. The token limits illustrate configuration structure; set them to actual model limits. tool_call: true declares support rather than adding tool capability.

Kilo uses uppercase URL in baseURL and {env:NIXAPI_API_KEY} for the key. These differ from Pi’s syntax. The official compatible-provider reference documents the provider structure.

4. Verify the model and tools

List configured models and launch Kilo:

kilo models
kilo

Start in a test project with a package.json. First request one short text reply, then ask:

Read package.json and summarize its scripts. Do not change files or execute commands.

Inspect the actual tool call and result. Once reading works, try a reversible edit and review the diff.

5. Editor setup and troubleshooting

For the Kilo editor extension, choose a custom OpenAI Compatible provider and enter the same endpoint, key, and model ID. Verify extension settings separately rather than assuming they synchronize with CLI configuration.

• Provider absent: check global config location, JSON syntax, and the model map.
• 401: check the environment in the launching terminal and whether configuration is trusted/global.
• Token-limit errors: use real limit.context and limit.output values.
• Text works but tools fail: verify streaming and tool calls for the chosen Chat Completions route before adjusting compatibility flags.