Unsloth puts the whole open-model workflow in one browser tab: pick a model, chat with it, train it, and ship it. Search a model hub, load a GGUF, and talk to it with web search, deep research, file uploads, and sandboxed Python or shell execution that asks before anything risky runs. Line up two models side by side to compare answers, or generate images and video with diffusion models. When a base model is not quite right, a five step wizard sets up fine-tuning with LoRA, QLoRA, FP8, or full training, and a live dashboard charts loss, gradient norm, step progress, and GPU memory. Data Recipes turn PDFs, CSVs, and Word files into question and answer pairs or instruction datasets without hand labeling. Finished models export to GGUF or safetensors for llama.cpp, vLLM, Ollama, and LM Studio. An access token screen exposes an OpenAI-compatible endpoint so Claude Code, Codex, and OpenCode can drive local models. Secure HTTPS links reach the interface from any device. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. Apache-2.0 and AGPL-3.0 licensed.