Run it yourself
Sovereign and private — AI on your own hardware. Your data never leaves your building, and you take on the upkeep. For the business where that trade is worth it.
AI you own, hardware and all
You run open-weight models on a machine you control — typically a computer with a good graphics card. The data never touches a third party. The models are downloadable, so there's no per-use bill.
Right for you if…
- Your data genuinely can't leave your building — regulated, confidential, or you just won't share it.
- You have the hardware (a decent GPU) and the appetite to maintain it.
- You value ownership — no vendor, no subscription, no one on the hook but you.
Not for you if you want a tool, not a project — this is the most hands-on of the four boxes.
Hardware once, electricity forever
The machine is the price of admission — a real GPU isn't cheap. After that, running costs are your power bill. No per-token fee, no monthly subscription. The hidden cost is your time: models need updating and the setup has a learning curve.
It's on you. Completely.
There's no SLA, no vendor on the hook, no support line. If it breaks or the model underperforms, that's your problem. Reliability here is whatever you make it — which is exactly the point, and exactly the trade.
Who plays here
- Ollama — the easiest way to run open-weight models locally; see the review.
- ComfyUI — for local image generation; see the review.
- Open-weight model families — the downloadable models themselves (e.g. Meta, Mistral, Qwen).
