Guides
Pick a model, then actually run it
Written for people who already use AI and want to run it on their own terms. Where a guide's advice turns on a particular model, it names the model and links to its page, which shows whether it runs on your machine and what it costs today.
New to this? Start with the primer, How to get started with AI models beyond ChatGPT, then pick the shelf for your job below.
Find a guide
Search results for “rent a GPU”
Browse all guides3 matches.
- DecidingRenting a GPU in the cloud: three ways to buy someone else's computeWhat most people asking for a GPU actually need is cheaper and duller, and this is the order to try the three options in.
- DecidingChoosing a graphics card for local models: VRAM beats everythingHow much memory decides what runs at all; how fast it moves decides the speed.
- How toBuild a second brain from the things you say out loudMeeting recordings into notes you can search: record, transcribe, distil, file and ask, all on a machine you already own.
Worth reading
How to · Speech and audioBuild a second brain from the things you say out loud
Catching what was said is the easy half, now that a good transcription model runs on a laptop. Two things go wrong anyway: people choose by reputation, when what decides the transcript is how a model copes with a room full of voices, and almost nobody gets as far as asking a question of the folder afterwards. This guide does both.
Updated 24 Sept
Buying a Mac for local models: Air, Pro or a Mac mini?The memory figure decides what you can run, and it is also the one people trade away to afford a better chip. This is the page with the ceilings on it: what each tier tops out at, which models that holds, and where the extra money stops buying anything you will notice.Who can see your prompts? Privacy, retention and where a provider livesWhen you rent inference, your prompts pass through someone else's machines. Whether that matters depends on three questions people blur into one — is it trained on, how long is it kept, and whose law applies — with different answers at the same provider.
Browse by what you want to do
Getting started
5 guides- How toHow to get started with AI models beyond ChatGPTThree ways to run a model, three models worth naming, and a first hour that settles more than a week of reading.
- How toRunning a model on your phoneYour phone will run a real language model offline. The ceiling is lower than the spec sheet suggests, and crossing it closes the app.
- How toRunning your first model locally, with Ollama or LM StudioTwo free tools turn this into an install and a download. The hard part is picking a model your memory can actually hold.
- DecidingWhen to consider running models locallyFour situations where a model on your own machine is the right answer, and the one sum that decides whether money is one of them.
- TroubleshootingWhy your local model is slow, and what to changeSlow local models have one law, a short list of levers, and a usual culprit that gives no warning at all.
Speech and audio
2 guides, plus 2 not written yet- How toBuild a second brain from the things you say out loudMeeting recordings into notes you can search: record, transcribe, distil, file and ask, all on a machine you already own.
- How toTurning speech into text on your own machineAn hour of audio, transcribed on the laptop you already own, with nothing leaving the room.
- Not written yetReading text aloud with a model on your own machineThe models that read text aloud, and which of them run without a graphics card.
- Not written yetChoosing a transcription model, and what it costs to runWhat the accuracy figures mean, which models run locally, and what a minute of hosted transcription costs.
Choosing hardware
3 guides- DecidingBuying a Mac for local models: Air, Pro or a Mac mini?Air, Pro, mini or Studio — the actual memory ceilings, what each one holds, and where the extra money stops buying anything you will use.
- DecidingChoosing a graphics card for local models: VRAM beats everythingHow much memory decides what runs at all; how fast it moves decides the speed.
- DecidingRenting a GPU in the cloud: three ways to buy someone else's computeWhat most people asking for a GPU actually need is cheaper and duller, and this is the order to try the three options in.
Running cheaper
1 guideBuilding with models
4 guides, plus 1 not written yet- How toEstimating what an AI feature will cost before you build itThe bill is arithmetic you can do before any code exists, and the surprise is that reading costs more than writing.
- How toPointing Claude Code, Codex or any tool at a different modelTwo settings, named for both tools, and a way to prove the request went where you sent it.
- How toUsing a small model as a classifierA component that reads a row and returns a label, sized by two hundred rows you labelled yourself.
- How toYou have tried one lab's API — how to try the restNearly everything speaks OpenAI's request shape, so trying another company is a base URL and a key.
- Not written yetRunning a model over a folder of your own imagesSorting, captioning or reading a batch of images without sending them anywhere.