Use the hardware note as a quick reality check before you pull a model.
Local AI, clearly catalogued
Find the right
open model
for the job.
A practical guide to capable open-weight models, with the context, license, and hardware notes you need before downloading.
59 modelscurated starting points
7 lanestext, code, vision & more
Local firsthardware notes included
The shortlist
Choose your starting point
Search by model or organisation, then save the ones worth testing.
A better first run
The model is only half the decision.
Context length, quantisation, and your available memory matter just as much as a benchmark score.
Open weights do not always mean the same usage rights. Read the model card.
A small local model that fits your workflow beats a large model you cannot run.
Field guide
Running a model locally
Ollama is the easiest starting point. llama.cpp offers more control, while LM Studio gives you a friendly desktop interface.
Weights are not the whole memory bill. Longer conversations need extra room for the KV cache, so start with a shorter context.
A good 4-bit quantisation is usually the sweet spot between quality and a model that actually fits on your machine.
Speed, tool use, and the quality of your prompts are personal. Use this catalogue to make a short list, then test it locally.