Unsloth
Unsloth is a desktop app and toolkit for running and training AI models locally. It supports LLMs, diffusion, embedding, and audio models, with features for fine-tuning, RL, and deployment. It is for developers who want to run or train models on their own hardware.
🎬 Unsloth Studio is insane… fine-tune any AI model locally · David Ondrej
✨ Key features
- Run and train LLMs, MLX, GGUF, diffusion, embedding, audio models.
- Fine-tune models 2x faster with 70% less VRAM.
- Supports RL, LoRA, QLoRA, full fine-tuning, pretraining, GRPO, DPO, FP8.
- Export models to GGUF, NVFP4, FP8 and more formats.
- Connect local models to agents like Claude Code and Codex.
- Access models via LAN or secure Cloudflare HTTPS.
🎯 Use cases
- Run local LLMs for chat and code generation.
- Fine-tune models on custom datasets.
- Train models with reinforcement learning (GRPO, DPO).
- Deploy models as OpenAI-compatible API endpoints.
- Connect local models to coding agents like Claude Code.
📦 Installation
🧰 Requirements: Works on Windows, Linux, WSL, and macOS. Supports NVIDIA, AMD, Intel GPUs, CPUs, and Vulkan. No API keys required for local use.
Unsloth Desktop (recommended)
Download from Unsloth or GitHub Releases.
Unsloth Studio (web UI)
macOS, Linux, WSL:
curl -fsSL https://unsloth.ai/install.sh | sh
Windows:
irm https://unsloth.ai/install.ps1 | iex
Unsloth Core (code-based)
Linux, WSL:
curl -LsSf https://astral.sh/uv/install.sh | sh
uv venv unsloth_env --python 3.13
source unsloth_env/bin/activate
uv pip install unsloth --torch-backend=auto
Windows:
winget install -e --id Python.Python.3.13
winget install --id=astral-sh.uv -e
uv venv unsloth_env --python 3.13
.\unsloth_env\Scripts\activate
uv pip install unsloth --torch-backend=auto
🚀 Usage
Unsloth Start
Connect Claude Code, Codex, and other agents to local models with one command:
unsloth start claude --model unsloth/Qwen3.8-27B-GGUF:UD-Q4_K_XL
Launch Unsloth Studio
unsloth studio
Docker
docker run -d -e JUPYTER_PASSWORD="mypassword" \
-p 8888:8888 -p 8000:8000 -p 2222:22 \
-v $(pwd)/work:/workspace/work \
--gpus all \
unsloth/unsloth
⚠️ Good to know
Server-side tools are on by default, so be careful when exposing Unsloth; keep your password safe or use --disable-tools.
❓ FAQ
What operating systems are supported?
Unsloth works on Windows, Linux, WSL, and macOS.
Can I use Unsloth with AMD or Intel GPUs?
Yes, Unsloth supports NVIDIA, AMD, Intel GPUs, CPUs, and the Vulkan backend.
How do I connect Claude Code to local models?
Use the command unsloth start claude to connect Claude Code to local models.
Can I access my models remotely?
Yes, you can use unsloth studio --secure to create a free Cloudflare link for global HTTPS access, or enable LAN access from settings.
📊 Repository
🤖 Overview, features, install steps and FAQ were generated from the project's README on Sep 4, 2026. Always check the original source before running commands.