Unsloth Studio has launched in beta as an open-source, no-code web interface for training and running AI models locally. The platform supports over 500 models including text, vision, TTS audio, and embedding models, with 2x faster training speeds using 70% less VRAM. Users can upload PDFs, CSVs, or JSON files to instantly start training on NVIDIA GPUs.

Built for accessibility, Unsloth Studio works on Mac, Windows, and Linux, with CPU-only chat inference available. The platform features Data Recipes for transforming documents into training datasets, real-time training monitoring, and model comparison tools. Users can export fine-tuned models to safetensors or GGUF formats for use with popular frameworks like llama.cpp and vLLM.

Privacy remains central to the design, with 100% offline operation and token-based authentication. While currently in beta with some limitations like precompiled binaries for faster installation, the team plans to add multi-GPU support, Apple Silicon/MLX training, and AMD/Intel compatibility. The platform represents a significant step toward making local AI development accessible to users without extensive technical expertise.