vibedonaldsvibedonalds.com
Unsloth Desktop screenshot

Unsloth Desktop

Free, open-source desktop app for Mac, Windows and Linux that runs and trains AI models on your own hardware.

What is Unsloth Desktop, in two sentences

Unsloth Desktop is a free, open-source app, in beta, for running and training AI models on your own hardware, available for macOS, Windows and Linux. It chats with local GGUF and MLX models, fine-tunes models without code, generates images and video, and connects Claude Code or Codex to local models.

About Unsloth Desktop

Unsloth Desktop is a Tauri app: you install it, choose a model and a quantization that fits your device in the Model hub, and start chatting with no setup. It relies on llama.cpp and Hugging Face for model support and on MLX and llama.cpp for CPU and GPU inference.

Models can use tools: Unsloth says its self-healing tool calls, which detect, repair and retry failures, are up to 50% more accurate, and models can run Bash and Python in a secure sandbox. Permission controls stop a model from accessing or editing your files or using the internet without your approval, unless you choose to give it direct access to your files.

The unsloth start command launches Claude Code, Codex, OpenCode, Hermes Agent, OpenClaw, VS Code's Copilot Chat and other agents against a local model, and an OpenAI-compatible API serves loaded models to other apps, scripts and SDKs. Models can be served on your local network or over HTTPS through a free Cloudflare tunnel, and the same chat interface also runs cloud models from OpenAI, Anthropic or a ChatGPT/Codex subscription.

Training needs no code: drop in a PDF, CSV or JSON for LoRA, full fine-tuning or pretraining, or train LoRA adapters for SDXL, FLUX.2, Qwen-Image and Z-Image on your own images. Media tools generate images with Qwen-Image-2.1, FLUX and Z-Image and video with Wan and LTX, inpaint, extend, upscale and edit images, and transcribe or generate audio locally.

Sources: unsloth.ai, /pricing, /download, /docs, /docs/desktop, /docs/integrations/unsloth-start, /docs/new/studio and /docs/models/qwen3.8-next, read 2026-10-26

Unsloth Desktop pros and cons

Pros

  • The docs promise no telemetry: the app detects your GPU type and device to know what works, and it can run entirely offline.
  • unsloth start sets the endpoint, API key, model and context length for each launch through session-scoped settings, without adding an Unsloth provider to the agent's normal configuration files.
  • Models you have already downloaded are found automatically, and you can point the app at custom folders.
  • Unsloth promises Day Zero support for new releases such as Qwen3.8, GLM and Gemma, and the model hub helps you pick the right quantization for your device.

Cons

  • Unsloth Desktop is still labelled Beta.
  • The docs say older hardware may not be well supported.
  • Web search, code execution and tool-call healing slow inference down; Unsloth says turning them off should bring speed in line with other llama.cpp apps.
  • The Codex integration works only with a GGUF model running through llama-server, not with the Unsloth transformers backend.

Tagged

  • Self-hostable
  • Supports Local Models
  • Web-based
  • Open Source

Unsloth Desktop alternatives

More from AI Coding Agents.

  1. 01
    Ondago

    Move AI coding agent sessions between computers, cloud, and phone

    →
  2. 02
    AlbertCode SWE

    AI coding agent that builds and tests code with your chosen model

    →
  3. 03
    BotGauge

    The trust layer for your agents

    →
  4. 04
    Orbi

    Lights-out software factory: file a GitHub Issue, get a release.

    →
  5. 05
    CloudKnife

    Cut your cloud bill, not your performance

    →
  6. 06
    PulseDev

    Click any UI bug on localhost and your AI agent fixes it

    →

Related reading

Reviews
No reviews yet — be the first to rate Unsloth Desktop.

Featured on Vibedonalds

Own Unsloth Desktop? Add this badge to your site to show you’re listed — and link back to your profile here.

Unsloth Desktop — Featured on Vibedonalds
<a href="https://vibedonalds.com/tools/unsloth-desktop" target="_blank" rel="noopener nofollow">
  <img src="https://vibedonalds.com/badge/featured-on-vibedonalds.svg" alt="Unsloth Desktop — Featured on Vibedonalds" width="240" height="60" loading="lazy" />
</a>

Frequently asked questions

What is Unsloth Desktop?
Unsloth Desktop is a free, open-source desktop app for Mac, Windows and Linux that runs and trains AI models on your own hardware.
Is Unsloth Desktop free?
Unsloth Desktop is free to use. There is no paid tier as of the last index update.
What platforms does Unsloth Desktop support?
Unsloth Desktop's own site lists it for macOS, Windows and Linux.
What are the alternatives to Unsloth Desktop?
Similar tools in AI Coding Agents on vibedonalds: Ondago, AlbertCode SWE, BotGauge, Orbi.
What are the downsides of Unsloth Desktop?
Unsloth Desktop is still labelled Beta. The docs say older hardware may not be well supported. Web search, code execution and tool-call healing slow inference down; Unsloth says turning them off should bring speed in line with other llama.cpp apps. The Codex integration works only with a GGUF model running through llama-server, not with the Unsloth transformers backend.
Is Unsloth Desktop free and open source?
Yes: the docs call Unsloth Desktop (Beta) a free, open-source app. The Unsloth Studio FAQ says the core Unsloth package is licensed under Apache 2.0, while optional components such as the Studio UI are AGPL-3.0. A separate pricing page lists Pro and Enterprise plans with faster training and multi-GPU support, with no price shown.
What is the difference between Unsloth Studio and Unsloth Desktop?
Unsloth Studio is an open-source, no-code web UI for training, running and exporting open models locally. Unsloth Desktop is the native app, and the Studio docs call it the easiest way to install Studio; Studio can also be installed manually with an install script or the Docker image.
Does Unsloth Desktop have web search?
Yes: the homepage promises unlimited, private and free web search inside the app. Normal search runs while the model is still thinking, and Deep Research plans first, searches for the best sources and produces a detailed report with citations.
How do I download and install Unsloth Desktop?
The download page offers a macOS build (42.7 MB), an Ubuntu/Debian build for Linux (44.3 MB) and a Windows 10/11 build (41.5 MB). The docs also give brew install --cask unsloth for Homebrew and winget install --id UnslothAI.Unsloth -e for WinGet.
Does Unsloth Desktop need a GPU?
No: the docs say Unsloth works on a wide variety of CPU and Mac setups, and supports NVIDIA, Intel, AMD and Mac GPUs and CPUs on Mac, Windows, Linux and WSL.