Free desktop app and open-source framework to run and fine-tune AI models locally on your own hardware.
Unsloth Desktop — API monitor and secure remote access
Overview
What is Unsloth?
Run and train models locally.
Unsloth is an open-source framework and desktop app for running and training AI models on your own local hardware. Unsloth Desktop lets you discover, download, and run nearly every open model locally with day-zero support for new releases, and it also trains models — the docs claim fine-tuning runs 2x faster with 70% less VRAM and no accuracy loss, across LLMs, diffusion, TTS, and embedding models. The unsloth start command connects coding agents such as Claude Code and Codex (plus VS Code, OpenCode, Hermes Agent, and OpenClaw) to your local models without changing your agent workflow, and an OpenAI-compatible API lets existing apps and SDKs connect through a familiar interface. Extras include unlimited private web search with deep research, self-healing tool calls with sandboxed Bash/Python code execution, and secure remote access to your models over HTTPS via a free Cloudflare tunnel. Native desktop apps are available for macOS, Windows, and Linux, with Docker and pip installs as alternatives. Unsloth's docs already cover NVIDIA's Nemotron 3.5 Bolt, making it one of the day-one runtimes for the October 2026 release.
Platforms and languages
Unsloth Availability
Platforms
Languages
Capabilities
Unsloth Key Features
Run models locally
Download and run nearly every open model locally, with day-zero support for new releases like NVIDIA Nemotron 3.5 Bolt.
Train and fine-tune
Fine-tune LLMs, diffusion, TTS, and embedding models — Unsloth reports 2x faster training with 70% less VRAM and no accuracy loss.
Connect coding agents
One command (`unsloth start claude` / `codex` / `vscode` / `opencode` / `hermes` / `openclaw`) routes agents to your local models.
OpenAI-compatible API
Serve models through an OpenAI-compatible API so existing apps, scripts, and SDKs connect through a familiar interface.
Search and code execution
Private, unlimited web search with deep research, plus self-healing tool calls that execute Bash and Python in a secure sandbox.
Plans and access
Unsloth Pricing
Free
Common questions
Unsloth FAQs
What is Unsloth?
Unsloth is an open-source framework and desktop app for running and training AI models locally on your own hardware. Unlike pure inference tools, it also fine-tunes models — claiming 2x faster training with 70% less VRAM — and connects coding agents to local models.
Which platforms does Unsloth support?
Unsloth Desktop has native apps for macOS, Windows, and Linux (also via Homebrew and WinGet), plus Docker and pip installs. A one-line curl/PowerShell installer is documented for macOS, Linux, WSL, and Windows.
Can Unsloth connect coding agents to local models?
Yes. Run `unsloth start claude` (or codex, vscode, opencode, hermes, openclaw) after loading a model, and the agent uses your local model instead of a cloud one. An OpenAI-compatible API is also exposed for other apps.
Does Unsloth support NVIDIA's Nemotron 3.5 models?
Yes — Unsloth was one of the runtimes NVIDIA named for its Nemotron 3.5 Bolt 74B-A9B launch (October 2026), and Unsloth's own docs already include a Nemotron 3.5 Bolt page.
Reviews
See what the community thinks and share your experience.
—
Based on 0 ratings
Rating distribution
Community reviews
No reviews
No written reviews yet.
