Unsloth

Unsloth is a desktop app for people who want to chat with models, train them on their own data, and create media on their own hardware. Its no-code interface brings these tasks into one workspace.

• Model chat and discovery: Run GGUF and MLX models, including Qwen3.8, DeepSeek-V4 and Gemma 4, and use the built-in model hub to find suitable downloads for your device.
• Training and datasets: Fine-tune language, diffusion, audio and embedding models with LoRA, QLoRA or full training, build datasets from PDFs, CSVs and DOCX files, and export models in formats including GGUF.
• Image and video creation: Generate and edit images with Qwen-Image-2.1 and FLUX, including inpainting and upscaling, or generate video with Wan and LTX.
• Coding agents and apps: Connect Claude Code, Codex and MCP to local models, or connect existing software through an OpenAI-compatible API; models can execute Bash and Python in a sandbox and retry failed tool calls.
• Web research: Search the web and produce reports with citations through features that need online access.

The software is free and open source under Apache 2.0. Desktop inference, training and media generation run locally on macOS, Windows and Linux. Hardware support includes CPUs, NVIDIA, AMD and Intel GPUs, multi-GPU setups and Vulkan. Unsloth Studio provides a self-hosted web UI with Docker support, while Unsloth Core is code-based. Optional Cloudflare tunnels provide remote access to models running on your hardware or on Colab.