RepoPicks Radar

Updated every day around 23:30 UTC

Guides

Unsloth

unslothai/unslothReviewed by AI

Checked by RepoPicks Radar on

What is Unsloth?

Unsloth is a desktop app (and a web interface and a Python package) that downloads and runs AI models on your own machine: chat models, image and video generators, speech models. The same app fine-tunes models, which the team says is about twice as fast and uses much less video memory. There are installers for Windows, macOS and Linux, and it works with NVIDIA, AMD and Intel cards or just the processor. Models are downloaded inside the app, so plan for several gigabytes of disk. The releases are marked beta.

Unsloth is free and open source (Apache-2.0) and can replace ChatGPT.

How easy is it to install Unsloth?

B · installer plus one simple step · Windows

  • Windows B
  • macOS B
  • Linux B

Every release ships ready installers (Windows .exe, macOS .dmg, Linux .deb and AppImage, around 20 to 170 MB), and there are Homebrew and Winget packages too, so on all three systems it is download, install and then pick and download a model inside the app: grade B. The command-line routes (install script, pip, Docker) exist but are not needed for the desktop app.

Unsloth on GitHub

Stars
79,292on Nov 10, 2026
Downloads
5,480,586472,004 this week
Growth
+271stars in 7 days

Stars over time

79,00079,10079,20079,300 Nov 4Nov 6Nov 8Nov 10 79,00079,10079,20079,300 Nov 4Nov 7Nov 10

Facts

Runs on
Windows, macOS, Linux
You need to install
Nothing extra
To build from source
Python, NVIDIA graphics card
Disk space
21 MB–62 MB · from the installer
Accounts or keys
Optional: a ChatGPT/Codex subscription or cloud model providers can be connected; not needed for local models
Cost
Free · Dual licence: the core package is Apache 2.0 and parts such as the Studio interface are AGPL-3.0, which matters if you want to redistribute a changed version or run it as a service.
License
Apache-2.0 (Permissive)
Last change on GitHub
Nov 10, 2026

Unsloth risk signals (7)

  • AttentionInstall command runs a script from the internet

    Install command pipes a remote script into a shell: curl -LsSf https://astral.sh/uv/install.sh | sh

    automatic check

  • Attentiontools enabled when exposed to network

    README.md, Remote HTTPS & LAN Access: "Server-side tools are on by default - so be careful! Keep your password safe, or use --disable-tools when exposing Unsloth." The same section offers a public Cloudflare HTTPS link and listening on 0.0.0.0.

    AI review

  • Infomodel can run commands and browse

    README.md features: "tool calling and code execution", "Browser use: Navigate sites, fill out forms, read PDFs, and edit files", and "Sandboxing: Restrict which files your model can access when it runs commands in a sandbox".

    AI review

  • Infosetup downloads extra prebuilt binaries

    The installer fetches extra prebuilt pieces: README.md says "Setup also installs audio.cpp (prebuilt by unslothai/audio.cpp)" and lets you pick a llama.cpp backend (UNSLOTH_LLAMA_CPP_BACKEND=vulkan/cpu/cuda/rocm); pyproject.toml ships node_prebuilt_pins.json and lemonade_prebuilt_pins.json.

    AI review

  • Infobeta releases

    The three latest releases are tagged as beta (v0.1.916-beta, v0.1.915-beta, v0.1.914-beta, November 2026).

    AI review

  • Infolocal server port open

    The app runs a local HTTP service: README.md explains UNSLOTH_DESKTOP_PORT, the default port 8888 and "If the selected port is busy, Desktop tries the next 20 ports".

    AI review

  • Infodual license agpl parts

    README.md License: "The core Unsloth package remains licensed under Apache 2.0, while certain optional components, such as the Unsloth Studio UI are licensed under... AGPL-3.0"; the repository carries both LICENSE (Apache) and COPYING (AGPL-3.0).

    AI review

Evidence is shown in its original language.

What Radar could not check about Unsloth

  • Real disk space: the installer is small, but the models you download inside the app take gigabytes.
  • The speed and memory claims ("2x faster", "70% less VRAM", "no accuracy loss") were not measured here.
  • How well it runs and trains on AMD, Intel, Apple Silicon or processor-only machines compared with NVIDIA cards.
  • How stable the beta desktop app is day to day.
  • What the app sends out when you use web search, the Cloudflare link or connected cloud providers.

Similar projects

The RepoPicks app

Coming soon: install in 1 click with the RepoPicks app.

Before you download

See the release page on GitHub