Unsloth

Unsloth is an open-source desktop application for running, training, and generating with AI models on your own hardware. It brings a no-code workspace, a model hub, local inference, image and video generation, private search, and a connection layer for coding agents into one app for Windows, macOS, and Linux. Because models and data stay on the machine, it suits people who want local control without assembling a stack of separate tools.

Key Features

• A free, open-source desktop app for Windows, macOS, and Linux with models stored locally.
• No-code training workflows for LoRA adapters, including 4-bit and 16-bit configurations.
• A built-in model hub that recommends quantizations for the available GPU memory.
• Image and video generation and editing with supported models and LoRA adapters.
• An OpenAI-compatible API plus a start command that connects coding agents to local models.
• Private web search, Deep Research, and sandboxed Bash and Python execution with retrying tool calls.
• Optional HTTPS tunneling so a running model can be reached from another device.

Who is it for

• Developers who want local model inference and a familiar API without a cloud account.
• Researchers and hobbyists training LoRA adapters on their own GPU.
• Creators who prefer to generate and edit images or video on local hardware.
• Teams handling sensitive experiments that should not leave their machines.

What can you do with Unsloth

• Download a model, load it, and start chatting from the desktop app.
• Train a LoRA adapter on your own data through a guided, no-code flow.
• Generate, transform, inpaint, extend, and upscale images, or produce short videos.
• Point Claude Code, Codex, or another agent at a local model through the start command.
• Run Deep Research with web search and citations while the model works.
• Serve a model over HTTPS and check a run from a phone or another computer.

How does Unsloth work

1. Install Unsloth Desktop and open the model hub.
2. Choose a model and a quantization that fits the available GPU memory.
3. Load the model, then chat, run tools, or start a training run from the workspace.
4. Connect an agent through the OpenAI-compatible endpoint or the start command.
5. Optionally open a tunnel and continue the session from another device.

Pricing

Unsloth publishes a free open-source edition and paid Pro and Enterprise editions. Pro and Enterprise are quoted after contact, so the current amounts are not listed on the pricing page. Check the official website for current pricing and plan details.

FAQ

Is Unsloth free to use?

The standard desktop app is open source and free to download. Pro and Enterprise editions add training speed, multi-GPU support, and enterprise features, and are quoted after contacting the team.

Can Unsloth run without an internet connection?

Models run locally after download. Some features, such as web search, Deep Research, model downloads, and remote access through a tunnel, need a network connection.

Do I need coding experience to train a model?

No. The desktop app includes no-code training workflows for LoRA adapters. Developers can still use the command line, the API, and the underlying open-source libraries.

Which kinds of models can I run?

Unsloth supports text, image, video, and audio models, with early support for new releases and a hub that offers quantized versions for different hardware.

Can I connect Claude Code or Codex?

Yes. The app exposes an OpenAI-compatible API and provides a start command that points coding agents at a local model without changing the agent workflow.

Does Unsloth send my data to a cloud service?

The app is built for local work, and models and files stay on your machine. Features that call external services, such as web search or tunneling, are optional.