Run AI on your Mac
— the easy way.
Local LLMs, ComfyUI, Claude Code and MCP. Practical guides for every Apple Silicon Mac, with no NVIDIA card and no cloud bill.
Tool · free
Can my Mac run it?
Pick your chip and memory. You get the models that actually fit — with an honest speed estimate and the trade-offs spelled out.
Run the checkStart here
All guides →Ollama on Mac: from install to your first model
Ten minutes, one terminal, no Python. The fastest path to a local model on Apple Silicon.
Local LLMsBest local LLM for a 16 GB Mac
What actually fits in 16 GB of unified memory — and where the ceiling really is.
ComfyUIComfyUI on Apple Silicon: the install that works
Desktop app, portable or manual — plus the MPS settings that stop it crawling.
Browse by topic
Local LLMs
Ollama, LM Studio and MLX — run models offline, keep your data on the machine.
ComfyUI
Node-based image and video generation on Apple Silicon, including MLX nodes.
Claude Code & MCP
Wire your Mac into an agent: configs, servers and workflows that hold up.
Build with AI
Ship a real website from your Mac — domains, hosting and deployment included.
Common questions
Can a Mac run large language models without a GPU?
Yes. Apple Silicon uses unified memory, so the CPU and GPU share the same pool, and Metal gives models direct access to it. A MacBook Air with 16 GB can run 8B-parameter models at 4-bit quantisation at a usable speed — no discrete GPU required.
How much memory do I need to run a local LLM on a Mac?
Plan for about 0.6 GB of memory per billion parameters at 4-bit quantisation, plus 6–9 GB for macOS and your apps. That means 8 GB handles 4B models, 16 GB handles 8B comfortably, and 24 GB and above opens up 14B models.
Is running AI locally on a Mac private?
Yes, if the model runs locally. Tools like Ollama download the weights once and then generate on your machine, so prompts and documents never leave the device. The only network traffic is the initial download.
Do I need ComfyUI to generate images locally on a Mac?
Not strictly — simpler apps exist for one-prompt-one-image work. ComfyUI is worth the learning curve when you need reproducible workflows, ControlNet, or video generation, and it supports both the Metal (MPS) and faster MLX paths on Apple Silicon.
Why this site exists
Most AI tutorials assume an NVIDIA GPU. On a Mac you have unified memory, Metal and MLX instead — different rules, different models, different limits. Every guide here is written for that machine, and every number is either measured or clearly labelled as an estimate. Read the methodology.