by Unsloth AI
Open-source desktop and web app for running, fine-tuning and serving open models locally, with agent and search tools.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
Unsloth is an open-source application for running and training open models on local hardware. It is offered as Unsloth Desktop (a native app for Windows, macOS and Linux), Unsloth Studio (a web UI) and Unsloth Core (the code-based library). It supports text, image, video, audio, embedding and decision models, with GGUF and MLX formats, and runs on NVIDIA, AMD and Intel GPUs, CPUs, Apple silicon and a Vulkan backend.
For training, it provides a no-code interface and a code library covering LoRA, QLoRA, full fine-tuning, pretraining, reinforcement learning (GRPO, DPO) and FP8. The project states 2× faster training with 70% less VRAM. Models can be exported to GGUF, NVFP4, FP8 and other formats, and datasets can be built from PDFs, CSVs and DOCX files with Data Recipes.
For inference, it includes a chat interface with tool calling, MCP support, sandboxed Bash and Python execution with user-set file and network permissions, private web search, deep research and browser use. The unsloth start command connects agents such as Claude Code, Codex, VS Code, OpenCode and others to local models, and an OpenAI-compatible API is exposed. Local models can be reached over LAN or through a Cloudflare HTTPS tunnel. It is free to use; performance depends on the user's own hardware.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Free open-source tool for local LLM fine-tuning that now also runs inference and has an agent UI similar to ChatGPT, with web search, MCP, tools and memory. It offers a no-code point-and-click interface for training and fine-tuning text, image and video models, supports remote access, and runs on Windows, Mac and Linux.
Closest neighbours by meaning, not just by category.
Model · Meta
Meta's open-weights 30B dense model for local agents, tuned for tool use and long tasks, Apache 2.0 licensed.
Model
Open-weights family of agentic coding models (9B to 397B) that learns to generate its own task-specific scaffolds.
Model
Open model family (9B dense, 35B MoE, 397B MoE) trained with a self-improvement loop for reasoning, coding and agentic tasks.
App · Z.ai
Free Hugging Face Space from Z.ai that runs GLM-powered vulnerability scans on submitted open-source GitHub repositories.
Model · Nvidia
NVIDIA's family of open models for agentic AI: reasoning, vision, speech, retrieval and safety models.
App · DeepSeek
DeepSeek's desktop agentic harness for running the DeepSeek models on coding projects, reportedly several in parallel.
