Guides
Know what to run, and how to start
Two kinds of page: situation guides that start from where you already are, and short concept explainers for the terms this site uses. Every guide hands off to the live catalogue rather than repeating numbers that go stale.
01
Start from your situation
Buying a Mac for local models: Air, Pro or a Mac mini?Capturing meetings and keeping the transcriptsChoosing a graphics card for local models: VRAM beats everythingEstimating what an AI feature will cost before you build itHow to get started with AI models beyond ChatGPTLeaving a flat-rate chat subscription for a metered APINot getting enough tokens per secondPointing Claude Code, Codex or any tool at a different modelRenting a GPU in the cloud — and why you probably want serverless insteadRunning a model on your phoneRunning your first model locally, with Ollama or LM StudioUsing a small model as a classifier, not a chat partnerWhen to consider running models locallyWho can see your prompts? Privacy, retention and where a provider livesYou have tried one lab's API — how to try the rest
02
The terms, explained
Two-minute explainers for the words the catalogue leans on. Each one says what the term changes about your decision, not just what it means.
VRAM and what fitsThe one number that decides whether a model runs on your machine at all.What are open weightsModels you can download and run yourself — with a licence that says what you may do with them.What is a context windowHow much a model can consider at once — the request, the history and the reply all share it.What is a model routerOne API key in front of many providers — ideal for tinkering, worth questioning in production.What is a tokenThe unit every AI price and every context limit is counted in.What is inferenceRunning a trained model to get answers — the part you pay for, in money or hardware.What is meant by "providers"Companies that run models on their hardware and sell access by the token.What is quantizationStoring a model's numbers more coarsely, so it fits in memory you actually have.
04