Chat with vision-language models.
Attach images, stream responses, and keep conversations local with models selected for practical Mac performance.
Native Apple Silicon app
MLXtra
Local AI creation for Mac, beyond chat.
Run chat, vision, image, speech, and music models on your Mac with a packaged MLX runtime, curated model downloads, and local output management.
One native workspace
Move from conversation to media without leaving the app.
MLXtra keeps the workflow direct: choose a model, prompt from the composer, and get generated text, images, speech, or music back inside the same conversation. Runtime setup, downloads, and generated files are handled by the app.
Workflows
Designed around the work, not a server setup.
Generate images from the same composer.
Use local image models, save generated files automatically, and open or export outputs when they are ready.
Create speech locally.
Generate spoken audio with local TTS models and keep outputs organized as app-generated media.
Make music with ACE-Step.
Prompt for instrumental or vocal music and receive local audio output inside the conversation.
Install curated models in Settings.
Download runtime components and models from the app, with size and hardware fit surfaced before you install.
Mac-first runtime
A local MLX stack without turning your Mac into a model server project.
Verified runtime
The app ships and updates a checked Apple Silicon runtime instead of relying on manual terminal setup.
Local outputs
Generated images, speech, and music are saved on your Mac and surfaced directly in conversation.
Practical catalog
Models are curated for realistic local use, with larger options shown only when the hardware fit makes sense.
Demo
See the native workflow in motion.
Model catalog
Curated local models for chat, media, and generation.
Qwen 3.5 9B
Vision language model with chat and image understanding support.
5.6 GB downloadQwen 3.5 2B
Smaller chat and vision model for lighter Apple Silicon Macs.
1.5 GB downloadFLUX.2-klein-4B
Local image generation and editing from the MLXtra composer.
15.0 GB downloadKokoro 82M 4-bit
Compact local speech generation for fast TTS workflows.
0.67 GB downloadACE-Step 1.5 Turbo
Prompt-based local music generation through an isolated runtime.
4.8 GB downloadQwen 3.6 35B A3B
Higher-capacity vision model shown when the Mac has enough memory.
20.4 GB downloadDownload
Run MLXtra on an Apple Silicon Mac.
First launch sets up the local runtime in the background. Pick a starter model, let downloads finish, then use the same app for chat, image, speech, and music workflows.
- macOS
- 14 Sonoma or later
- Chip
- Apple Silicon
- Runtime
- MLXtra runtime 0.1.5
- Catalog
- 2026.06.01.1