Overview

Unsloth Desktop is a free, open-source local AI workspace for running, training, and connecting text, image, video, and audio models on your own hardware. A built-in model hub picks the right quantization for your device, with Day Zero support for new releases like Qwen3.8, GLM, Gemma, and DeepSeek. Beyond chat, it does no-code fine-tuning — LoRA, full fine-tuning, and pretraining from a PDF, CSV, or JSON — at roughly 2x speed with 70% less VRAM. It generates and edits images with FLUX, Z-Image, MiniMax-H3, and LoRA adapters, and video with Wan and LTX. Models run code in a sandboxed Bash and Python environment with self-healing tool calls that detect, repair, and retry failures. An OpenAI-compatible API and the `unsloth start` command connect Claude Code, Codex, and other agents to your local models, and a free Cloudflare tunnel serves them over HTTPS from anywhere. Currently in beta.

Pricing: Free and open source

Architecture: Apple Silicon, Intel

Key Features

Run local text, image, video, and audio models with no setup
Built-in model hub with per-device quantization picks and Day Zero model support
No-code fine-tuning — LoRA, full fine-tuning, and pretraining from PDF, CSV, or JSON
2x faster training with 70% less VRAM and no accuracy loss
Image generation and editing with FLUX, Z-Image, MiniMax-H3, and LoRA adapters
Video generation with Wan and LTX
Sandboxed Bash and Python execution with self-healing tool calls
OpenAI-compatible API for existing apps, scripts, and SDKs
Connect Claude Code and Codex to local models with `unsloth start`
Unlimited private web search with Deep Research and cited reports
Serve models over HTTPS through a free Cloudflare tunnel
Runs fully offline on your own hardware