Editorial policy
This page explains how we keep what we publish honest, and independent of how we make money. It is a promise we intend to keep, and one you can hold us to.
Where we stand
We think running AI on your own hardware matters, for privacy, for autonomy, and for the health of an open ecosystem. That view shapes what we choose to cover. It does not shape how we report it.
We are not against cloud AI. Cloud models are often more capable, always easier to start with, and frequently the right choice. We say so where it is true. If our content leads you to conclude that an API is the better answer for you, we have done our job. Local AI is also often slower, takes more effort, and needs expensive hardware, and model quality at consumer scale still trails the frontier. We would rather tell you that plainly than sell you something.
Editorial independence
Our news coverage, model rankings, guides, and the strengths and weaknesses we list for any tool are decided on the merits alone. No advertiser, sponsor, or affiliate partner has any influence over them, and none is offered any. The three ways we make money, described below, are kept strictly separate from that editorial judgement.
Featured placements
Solutions and hardware can hold paid featured status. When they do:
- They carry a visible Featured badge.
- They appear in a distinct, clearly labelled section above the organic results, never mixed in with them.
- A featured entry still lists its genuine weaknesses, written to the same standard as any other. If we cannot write an honest weaknesses list for something, we do not accept the placement.
Featured status buys visibility, and nothing else. It never affects our news coverage, our model rankings, the content of our guides, or the assessment on a product's own page. If a paid placement would require us to soften an honest assessment, we decline it.
Affiliate links
Some links, mainly to hardware, are affiliate links, which means we may earn a small commission if you buy through them at no extra cost to you. Where a page contains one, it is marked inline with a small (affiliate) label, and the page carries a short disclosure line.
Recommendations are made on merit, and the affiliate link follows the recommendation, never the other way round. If the best option for a use case has no affiliate programme, it is still the option we recommend.
Advertising
We run display advertising on some pages. We deliberately keep it off our interactive tools, the hardware matrix and the cost calculator, as well as the glossary and this policy. Those tools are the site's credibility anchors, and we would rather protect them than monetise them.
Benchmarks and specifications
We do not run our own benchmarks. Where we cite them, we attribute the source and give the date, because benchmarks go stale, get re-run, and get gamed. We treat figures from a releasing organisation as claims until they are independently confirmed. VRAM figures and cost projections are estimates, and we label them as such wherever they appear. Fitting a model in memory is not the same as running it well, and we try never to blur the two.
Corrections
We get things wrong sometimes. When we do, we correct the error in place and add a short, dated correction note at the foot of the article or entry. We do not quietly edit and move on, and we try not to be defensive about it. If you spot a mistake, particularly a wrong specification or licence classification, please tell us.
Our own products
Looking ahead: Ragtag, a local-first desktop RAG application, is related to the people behind this site. If and when it appears in the solutions catalogue, it will be listed under exactly the same criteria as every other entry, with an honest weaknesses list, and it will receive no ranking or placement advantage over comparable tools. Its relationship to the site will be disclosed on its own entry as well as here. We are noting it now so the disclosure is on record before the fact.