Continue.dev

Continue is an open-source coding-assistant extension for VS Code and JetBrains. It reads its model list from ~/.continue/config.json, and any entry with an OpenAI-compatible apiBase can point at a local rapid-mlx server — chat, edit, and autocomplete all stay on your Mac.

Install

Install the Continue extension from the VS Code marketplace or the JetBrains plugin repository, then open it once so it creates ~/.continue/.

Config — one shot

$ rapid-mlx launch continue-dev

This appends a model entry named rapid-mlx to ~/.continue/config.json and promotes it to the default — your existing Continue models are kept, and re-running updates the entry in place instead of stacking copies. What it writes:

{
  "title":   "rapid-mlx",
  "provider": "openai",
  "model":   "default",
  "apiBase": "http://127.0.0.1:8000/v1",
  "apiKey":  "sk-noop"
}

Prefer to edit by hand? Paste the same object into the top-level models array of ~/.continue/config.json. On Continue's newer YAML config, the equivalent block goes under models: in ~/.continue/config.yaml with the same five keys (name instead of title).

Run

$ rapid-mlx serve qwen3-coder-30b-4bit

Open the Continue panel and pick rapid-mlx from the model dropdown. model: "default" always resolves to whatever alias the server booted, so switching models is a server-side decision — no config edits.

Gotchas

• The API key must be non-empty. Continue refuses an entry with a blank apiKey even though rapid-mlx doesn't check it — that's why the launcher writes sk-noop.
• Autocomplete wants a small, fast model. Continue's tab-autocomplete is configured separately (tabAutocompleteModel) and fires on every pause — point it at a 4B-class alias (e.g. qwen3-4b-instruct-2507-4bit) rather than your chat model, or leave it on Continue's default.
• JSON vs YAML config. Newer Continue versions read config.yaml alongside config.json; the launcher targets the JSON file, which Continue still honors. If you've migrated fully to YAML, add the entry there instead.

See also