Run LLMs on your Mac.
Download any open-source AI model and run it on Apple machines.
No cloud, no subscriptions, complete privacy — with best UI.
Download any open-source AI model and run it on Apple machines.
No cloud, no subscriptions, complete privacy — with best UI.
Organize conversations by topic, client, or workflow. Every thread stays exactly where you put it.
Run multi-step research sessions locally. No rate limits. No token caps. Go as deep as you need to.
Pull live web data into any conversation. Local models, real-time answers.
Generate images on-device with open-source diffusion models. No API keys. No per-image fees.
Create documents, code, and other outputs in a side-by-side workspace. Edit without leaving the conversation.
Wire your AI to calendar, email, browser, and files. It doesn’t just think — it acts.
The full HuggingFace catalog, searchable inside MacLLMs. Search any model, download and load it — no terminal, no config, no commands.
MacLLMs reads your chip, RAM, and GPU on first launch. Every model in the Hub shows compatibility, memory needs, and speed estimates for your exact hardware. You'll never download something that won't run.
Select a base model. Drop in your dataset. Hit train. Watch loss curves, checkpoints, and metrics update live — through a visual dashboard built for focus, not frustration.
Turn raw documents into training-ready datasets. Generate synthetic instruction-response pairs, clean messy files, validate schemas, and export in ChatML, Alpaca, or ShareGPT format — all without writing a single script.
“A model trained on just 1,000 curated examples outperformed one trained on 52,000 — and matched GPT-4 in 43% of human evaluations.”
SOURCE: LIMA PAPER · NEURIPS 2023Import raw documents, CSVs, PDFs, or paste text. Or start from scratch with synthetic generation.
Deduplicate, fix encoding, normalize formats. Auto-detect and repair schema issues. Score quality per row.
Generate instruction-response pairs using teacher models. Augment sparse datasets. Build SFT-ready conversations from raw knowledge.
Train-ready file
Open-source models. On-device inference. One purchase. No recurring fees, no data collection, no strings.
Download for Mac