
About
What's New in Juno
4.4.1
October 27, 2026
What's New in Juno 4.4: AI & ML ON-DEVICE – Run language models, generate images and train neural networks on your iPhone or iPad, from your own Python code: llama-cpp-python, MLX, mlx-lm and Keras now ship with Juno. – Five ready-to-run projects in Home's new AI & ML section: chat with a local LLM, Stable Diffusion image generation, Keras training, RAG over your own notes, and LoRA fine-tuning. Models download on the first run and work offline after that. MLX, KERAS AND STABLE DIFFUSION – Apple's MLX array framework and mlx-lm are built in: run language models and fine-tune them with LoRA adapters, right on your device. – Build and train neural networks with Keras on a fast MLX backend — the same Keras code you would write on a laptop. – Generate images from a text description with Stable Diffusion running on MLX; a compact model downloads first, and larger ones are one line away. RUN LLMS LOCALLY – Chat with a GGUF language model through llama-cpp-python, the Python bindings for llama.cpp, with replies streamed word by word. Start with the Qwen3 download, or drop in a GGUF file of your own. – Ask questions about your own notes: a small embedding model finds the relevant passages and a local LLM answers from those passages only, so nothing leaves your device. HUGGING FACE BUILT IN – huggingface_hub, transformers, tokenizers and safetensors now ship with Juno — nothing to install, just import them. Downloaded models are kept in Files, in Juno's huggingface folder. MORE – New built-in packages, including pymunk for 2D physics and audioop-lts. – Leaving the REPL now keeps your session, so you can pick up where you left off. – Notebook settings such as Wrap Lines and Show Line Numbers now apply to newly opened notebooks without a relaunch. – Security updates: refreshed trusted certificates and a fix for a vulnerability in the bundled Brotli decoder (CVE-2026-111497). – A range of stability and reliability fixes.
More





