Unsloth

Unsloth is an open-source platform designed for training, running, and exporting open models locally. It provides a unified no-code web UI that operates entirely offline on Mac and Windows devices. The tool is aimed at developers and users looking to run and fine-tune models directly on local hardware.

• Run and train models locally using a 100 percent offline Mac and Windows application.
• No-code interface for auto-creating datasets from PDF, CSV, and JSON documents.
• Custom training kernels supporting optimized training for LoRA, FP8, FFT, PT, and over 500 models.
• Unlimited tool-calling and web search capabilities with sandboxed execution of Bash and Python programs.
• Data Recipes workflow to transform structured or unstructured documents into usable datasets.
• Model export options supporting GGUF and Safetensors formats.

Pros

• Allows complete offline operation on local Mac and Windows hardware for privacy and independence.
• Provides built-in dataset creation from multiple file formats like PDF, CSV, and JSON without writing code.
• Supports model export to formats compatible with external runtimes like llama.cpp, vLLM, and Ollama.
• Offers side-by-side model comparison alongside the ability to upload images, documents, audio, and code files.

Cons

• Buyers should check specific hardware requirements, as local execution depends heavily on local device capabilities.
• Multi-GPU support is noted as still in the works on the provided page text.
• Specific pricing tiers beyond the free version are not published on the page.

Unsloth fits developers and individuals wanting to run, train, and fine-tune open models locally and offline without writing code. It is suitable for users who need to process documents into datasets, execute sandboxed code computations, and export models to local runtimes. It is a poor fit for organizations requiring managed cloud hosting out-of-the-box without local hardware deployment.

The product is described as a fully free, open-source version, with free fine-tuning also available on Google Colab or Kaggle Notebooks. No other paid tiers or pricing details are published on the page.

Unlike standard cloud-based model platforms, Unsloth operates completely offline on local Mac and Windows devices while providing a no-code web UI. It combines custom training kernels with built-in dataset creation workflows and sandboxed program execution like Bash and Python directly inside the interface. Users can execute unlimited web searches and run models locally without relying entirely on external API providers.

GGUF, Safetensors, llama.cpp, vLLM, Ollama, Google Colab, Kaggle Notebooks, PDF, CSV, JSON, and OpenAI compatible APIs.