Unsloth launches first local AI desktop app with 2x speed boost
Run LLMs locally on Mac/Windows/Linux with 2x speed and 70% less VRAM...
Unsloth has launched its first desktop application, Unsloth Desktop, which allows users to run and train AI models entirely locally. This open-source software supports multiple platforms, including Mac, Windows, and Linux, and is compatible with frameworks like MLX (Apple’s ML acceleration library), diffusion models for image/video generation, audio models, and GGUF (a format for quantized models). The app supports popular models such as MiniMax-H3 and Muse Glimmer, with upcoming support for Qwen 3.8. It also enables integration with tools like Claude Code and Codex, offering features like 50% more accurate self-healing tool calls and sandboxed code execution.
The desktop app accelerates model training by 2x while reducing VRAM usage by 70%, making it far more efficient than traditional setups. It includes advanced features like private web search, deep research, RAG (retrieval-augmented generation), MCP (model context protocol), and exports to formats such as NVFP4 and GGUF. Users can interact with the app via an OpenAI-compatible API and even deploy models securely using Cloudflare HTTPS, with no telemetry or data collection. Unsloth Desktop is available now for free on unsloth.ai and GitHub.
- First open-source desktop app for local LLM training/inference, supporting Mac/Windows/Linux
- 2x faster training with 70% less VRAM, supports GGUF, MLX, diffusion models, and integrates with Claude Code
- No telemetry, secure deployments via Cloudflare HTTPS, and OpenAI-compatible API
Why It Matters
Unsloth Desktop empowers developers to run powerful AI models locally with minimal hardware overhead, enhancing privacy and speed for AI workflows.