Repolode
Submit
unslothai

unslothai/unsloth

Open-source desktop app and web UI for running, fine-tuning and exporting LLMs and diffusion models locally, plus a code-based training library.

Open-source alternative to:LM StudioOllama

View on GitHubWebsite
Stars
78,299
Forks
7,246
Open issues
325
Last commit
3 hours ago

What is it?

Unsloth is a local tool for running and training AI models. It ships as Unsloth Desktop, a native app for Windows, macOS and Linux, and as Unsloth Studio, a web UI launched with the unsloth studio command. Unsloth Core is the code-based Python package for fine-tuning in scripts and notebooks.

On the run side, it supports GGUF and MLX models, diffusion image and video models, and provides an OpenAI-compatible API. It can also connect local models to agents such as Claude Code, Codex and MCP tools, and can be reached over LAN or through a Cloudflare HTTPS link.

On the training side, it covers LoRA, QLoRA, full fine-tuning, reinforcement learning methods such as GRPO and DPO, and export to formats including GGUF and FP8. The project is licensed under Apache 2.0 for the core package and AGPL-3.0 for the Studio UI.

Key features

  • 01Runs and trains LLMs, diffusion, audio and embedding models on Windows, Linux, WSL and macOS.
  • 02Supports NVIDIA, AMD and Intel GPUs, CPUs and the Vulkan backend.
  • 03Serves models through an OpenAI-compatible API.
  • 04Connects local models to Claude Code, Codex and MCP tools.
  • 05Supports LoRA, QLoRA, full fine-tuning, GRPO and DPO.
  • 06Exports trained models to GGUF, FP8 and other formats.
  • 07Builds datasets from PDFs, CSVs and DOCX files.

Use cases

  • Running open-weight models locally on a laptop or workstation without sending data to a cloud service.
  • Fine-tuning a small LLM on a single consumer GPU with lower VRAM use.
  • Connecting a local model to coding agents such as Claude Code or Codex.
  • Training diffusion or TTS models and exporting them for local deployment.

Quick start

  1. 1

    Download the Unsloth Desktop app for your OS from the releases page

  2. 2

    Install on macOS with Homebrew

    brew install --cask unsloth
  3. 3

    Launch the Unsloth Studio web UI

    unsloth studio

Things to know

  • Server-side tools are on by default, so exposing Unsloth over the network needs a strong password or the --disable-tools flag.
  • Some install paths, such as the Docker GPU setup, depend on NVIDIA or AMD hardware.

Topics

#llm#fine-tuning#local-llm#gguf#diffusion#reinforcement-learning#openai-compatible#self-hosted#agent#ai

FAQ

What is unsloth?
Open-source desktop app and web UI for running, fine-tuning and exporting LLMs and diffusion models locally, plus a code-based training library. Unsloth is a local tool for running and training AI models. It ships as Unsloth Desktop, a native app for Windows, macOS and Linux, and as Unsloth Studio, a web UI launched with the unsloth studio command. Unsloth Core is the code-based Python package for fine-tuning in scripts and notebooks.
Is unsloth free to use? What license does it have?
unsloth is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is unsloth still maintained?
Actively maintained. The last commit to unsloth was on Oct 20, 2026.
What can I use unsloth for?
Running open-weight models locally on a laptop or workstation without sending data to a cloud service. Fine-tuning a small LLM on a single consumer GPU with lower VRAM use. Connecting a local model to coding agents such as Claude Code or Codex. Training diffusion or TTS models and exporting them for local deployment.
How do I install unsloth?
Download the Unsloth Desktop app for your OS from the releases page Install on macOS with Homebrew: brew install --cask unsloth Launch the Unsloth Studio web UI: unsloth studio
Where can I get unsloth?
The source code and installation instructions are on GitHub: https://github.com/unslothai/unsloth

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects