Unsloth
Free tierRun and Train Models Locally
Free tier available·All audiences·API available·Open source
Key strengths
Local-first AI model running and trainingOpen-source and free desktop appOpenAI-compatible API for agent integrationSignificant VRAM reduction (up to 90% on Enterprise)Significant training speed improvements (up to 32x on Enterprise)Day Zero support for latest modelsImage and video generation supportSelf-healing tool calls for improved accuracyCross-platform (Mac, Windows, Linux)
Free tier + paid plans
Founded 2023
Self-hostable
No ratings yet
Unsloth — Technical Overview
Unsloth is an open-source, local-first framework for running and fine-tuning large language models (LLMs) and diffusion models on consumer and professional hardware. Founded in 2023, it has accumulated over 77,000 GitHub stars, reflecting strong adoption in the developer and research community.
Architecture & Performance
- Delivers up to 2x training speed and 60% VRAM reduction on the free tier via 4-bit and 16-bit LoRA fine-tuning on a single GPU
- Pro plan scales to 2.5x faster training, 80% VRAM reduction, and multi-GPU support (up to 8 GPUs)
- Enterprise plan reaches 32x faster training, up to 30% accuracy improvements, 5x faster inference, and multi-node distributed training
API & Integrations
- Exposes an OpenAI-compatible API, enabling drop-in compatibility with agent frameworks and tooling
- Native integrations with Claude Code, Codex, Cloudflare tunnel, GGUF, and MLX
- Supports self-healing tool calls to improve function-calling reliability in agentic pipelines
Deployment
- Fully self-hostable; cross-platform support for Mac, Windows, and Linux
- Free desktop app available under open-source license
- Day Zero support for the latest model releases
- Supports image and video generation workloads in addition to text models