Kilo Code
Kilo CLI can use a local Rapid-MLX model through its OpenAI-compatible provider.
Quick start
$ npm install -g @kilocode/cli $ rapid-mlx serve qwen3.5-4b-4bit $ rapid-mlx agents kilo-code --setup $ kilo run --pure --auto "Explain this project"
Setup detects the served model and context limit, then merges the provider into ~/.config/kilo/kilo.json. For a different port, add --base-url http://127.0.0.1:18730/v1. Use --dry-run to preview. The local key is not-needed when the server has no authentication.
Config and editor
The provider uses @ai-sdk/openai-compatible, a baseURL ending in /v1, and provider/model selection. Kilo CLI 1.0 and later use ~/.config/kilo/kilo.json[c]; they no longer search .opencode. Project kilo.json[c] can override global settings.
In the VS Code extension, open Providers → Custom provider. Give it an ID such as rapid-mlx, select the OpenAI Compatible protocol, enter the local /v1 URL and model ID, then submit. Use not-needed as the key for an unprotected server. The extension's selected model may need to be switched after CLI setup.
Check the connection
kilo models rapid-mlx should list the configured model. rapid-mlx agents kilo-code --test runs wire and CLI checks. A tool call counts only when the requested file or shell action actually happens.