Practical local AI for real hardware

Find the local AI setup your computer can actually run.

Tell us what computer you have and what you want to do. You will get a realistic model class, app, and setup path before you waste time downloading the wrong thing.

Check my setup Learn how memory fit works

Example result

16GB Apple Silicon Mac

Start with a 7B–8B model in a common 4-bit build. Use LM Studio for the easiest graphical setup, or Ollama when you want a local runtime and API. Keep context modest at first.

Model class: 7B–8B, 4-bit
First app: LM Studio
Bottleneck: Unified-memory headroom

Compare LM Studio and Ollama →

Hardware Fit Lab

Find a sensible first setup.

Choose the machine you have and the job you need done. The result gives you a model class, app, and first next step.

Computer

Memory

Graphics

Goal

I do not know my specs

Mac: Open the Apple menu, choose About This Mac, and note the chip and memory.

Windows: Open Settings → System → About for installed RAM. Open Task Manager → Performance → GPU for the GPU name and dedicated GPU memory.

Linux: Use the system information panel, or run free -h for memory and lspci | grep -E "VGA|3D" for graphics.

Start with the job

Four useful paths. No catalog archaeology.

Pick the outcome you need. The deeper model, hardware, and runtime directories remain available when you need the detail.

I want the simplest first setup

Install one desktop app, choose one sensible model, and get a successful first response before touching advanced settings.

Use the graphical setup path

I want to chat with documents

Set up extraction, embeddings, retrieval, and a model that fits your machine instead of treating PDF chat as one magic feature.

Build a document workflow

I need the right model for my machine

Choose by memory, task, and runtime support instead of parameter-count hype or the biggest download that barely loads.

See the curated model shortlist

Something is broken

Start with the symptom: a model will not load, generation is slow, the GPU is missing, apps cannot connect, or document answers are bad.

Diagnose the failure

Common decisions

The questions that usually decide the setup.

8GB, 16GB, or 32GB?See what each memory tier changes.Ollama or LM Studio?Choose the terminal/API path or the graphical desktop path.Is the workflow actually local?Map the model, embeddings, storage, network, and tools.

Need the deeper reference layer?

Browse model details, hardware archetypes, runtimes, comparisons, and machine-readable data after you know what decision you are trying to make.

Explore data