mlx-runner v1.1.0 – Local Image Generation

mlx-runner v1.1.0 adds fully local image generation on Apple Silicon.

Highlights

• Added the Create Image preview in the chat sidebar.
• Integrated the 4-bit MLX checkpoint toxicdog/Qwen-Image-2.1-MLX.
• Added one-click image-model downloads from Create Image and Settings.
• Reused the existing local model storage, download progress, status handling, and recovery flow.
• Added lazy loading and unloading so unified memory can be reclaimed.
• Added prompt, dimensions, steps, seed, generation timing, PNG saving, and validation.
• Improved first-run Apple Silicon/Python checks, web-port handling, and dependency refresh detection.
• Migrated FastAPI startup to the modern lifespan API.

Everything remains local after download. No cloud image fallback is used.

MLX Runner v0.1.0 – Initial Public Release

MLX Runner v0.1.0

Initial public release of MLX Runner — a lightweight, MLX-native local LLM runtime for Apple Silicon.

Why MLX Runner?

Running local models on macOS often means:

• Dependency issues
• Terminal-only workflows
• Manual configuration
• Complex environment setup

MLX Runner aims to remove that friction.

Goal:
Download → Run one command → Start local models.

---

Features

• Native MLX backend
• Simple local GUI
• Fully offline execution
• Minimal setup
• Designed for Apple Silicon

---

Intended Use

• Run local MLX-compatible models
• Experiment with reasoning workflows
• Foundation layer for extended tooling

---

Notes

This is an early release (v0.1.0).

Expect:

• Rough edges
• UI improvements
• Performance tuning
• Extended tooling support

Feedback, issues and contributions are welcome.