oMLX

Use locally running oMLX models as AI providers in Raycast. Your models appear natively in AI Chat, Quick AI, and AI Commands — no cloud, no API costs.

Requirements

• macOS with Raycast installed
• oMLX installed and launched at least once
• At least one downloaded model

Setup

1. Install oMLX and launch it once.
2. Install this extension.
3. Enter your oMLX API key (found in ~/.omlx/settings.json under auth.api_key).

AI Chat

Type @omlx in Raycast AI Chat to manage models conversationally — load, eject, search, download, check stats, and control the server without leaving the chat.

Commands

Manage Models

Browse loaded, available, and helper models. Load, eject, pin, favorite, or delete models. View model details and per-model performance stats.

Download Model

Search HuggingFace and ModelScope for models to download. Filter to MLX-only models. Browse trending and popular models.

Manage Downloads

Track HuggingFace and ModelScope download progress with live updates. Cancel, retry, or remove downloads; ModelScope tasks are labeled by source. If one source is unavailable, the other source's downloads stay visible with an inline warning and automatic recovery checks.

Serving Stats

View prompt processing speed, token generation speed, cache efficiency, memory usage, and active models. Switch between session and all-time stats. View server logs.

Check for Updates

Check if a new version of oMLX is available.

Start/Stop Server

Start, stop, or restart the oMLX server.

Open Web Dashboard

Open the oMLX web dashboard in your browser.

Development

Install dependencies with npm ci, then run npm run dev to load the extension in Raycast. Run npm test for mocked streaming, search-race, update-notification, and download-routing regressions; these tests do not require a running oMLX server. Validate changes with npx tsc --noEmit, npm run lint, and npm run build. Run npx ray evals separately for conversational tool behavior.