Unsloth
Local UI to run and train LLMs and diffusion models on your own hardware -- 2x faster, 70% less VRAM.
About Unsloth
Unsloth is a desktop application and Python library for running, fine-tuning, and training large language models and diffusion models locally. It supports 500+ models including Qwen3, DeepSeek-V4, Gemma 4, Kimi K3, FLUX, MiniMax-H3, and more. Unsloth uses custom Triton kernels and optimized fine-tuning techniques (LoRA, QLoRA, full fine-tuning, GRPO reinforcement learning) to deliver up to 2x faster training with 70% less VRAM usage.
Unsloth Studio provides a web UI for chat, model comparison, API endpoints (OpenAI/Anthropic compatible), and MCP tool integration. Unsloth Start connects local models to AI coding agents such as Claude Code, Codex, OpenCode, and Hermes Agent. Deployable via Docker, native desktop apps (Windows/macOS/Linux), or pip install.
Key Features
- Run and train 500+ LLMs locally with 2x faster, 70% less VRAM usage
- Fine-tuning with LoRA, QLoRA, full training, RL, and long-context support to 500K tokens
- Web UI with chat, model comparison, API endpoints, and MCP integration
- Connect local models to Claude Code, Codex, OpenCode, and Hermes Agent
Self-Hosting Notes
Docker image: unsloth/unsloth. GPU required for training (NVIDIA, AMD, Apple Silicon). Desktop app available for Windows/macOS/Linux. pip install also available. Dual-licensed Apache 2.0 and AGPL 3.0.
More in AI, Automation, & Workflow Orchestration
Activepieces
Business automation tool focusing on marketing, sales, and operations flows.
Appwrite
Backend platform for web and mobile Firebase open-source alternative.
Budibase
Low-code platform for building business apps and internal tools.
Dify
Comprehensive LLM application development platform for building AI plug-ins and assistants.