Skip to content

Privacy

RigCheck has no accounts and sets no cookies of its own. Apart from emails you send us, the only personal data we keep is what people who sign up for email alerts give us: an address and the kinds of hardware they picked, and what people put in a measurement report: the report as sent, including its pasted log, which may contain personal details if they were not removed, and an optional name to credit. The checkers and calculators run entirely in your browser; they never send the hardware you select to us. Only two forms send what you enter, when you press their button: an email-alert form (the address and the hardware categories shown in it) and the measurement form (the report). A "copy link" button copies a link with your selections after the # so you can share them; browsers do not send that part to any server. When you open such a link, the page reads the selections and then removes them from its address. (Share links made before 7 October 2026 put them after a ? instead, which browsers do send when the page is requested; the page moves them behind the # as soon as it opens.)

What we measure, and why

None of them receives the hardware you pick. Details for each follow.

What stays in your browser

The home page finder remembers the last graphics card and RAM you picked in local storage on our private tool origin, tools.localrigcheck.com, so they are there on your next visit. It is never sent anywhere; clearing this site's data removes it. Its "copy link" button copies a link to that unrecorded origin with the selection after the # without changing the page address, like the checkers' share links. Old shared links to the main site are sent directly to the private checker without starting Clarity. Historical finder selections on the main origin are deleted without being read or transferred. Ordinary finder links use clean addresses; to prefill a checker, a one-use selection is kept in session storage for up to one minute and consumed when a checker opens. It stays in your browser. Links opened in a new tab or with storage blocked use the checker defaults.

When nothing is selected yet, the finder asks your browser which graphics chip it draws with (the WebGL renderer name, such as "NVIDIA GeForce RTX 4070") and offers that card as a suggestion. The name is matched against the finder's own list inside your browser; it is not stored, not counted and not sent anywhere, and nothing is selected until you click the suggestion. Browsers do not report graphics memory, and on laptops with two graphics chips they may report the integrated one, so check the suggestion before using it.

If you email us a correction or a measurement, we use your address only to reply. Measurements can also be sent through the measurement form, described below. A measurement is published, only with your agreement to share it under CC BY 4.0, with its hardware, settings and numbers, without your name unless you ask for it.

We use privacy-friendly, cookieless Cloudflare Web Analytics to count page views. It does not track you across sites or store personal data on your device.

Google Analytics

We also use Google Analytics 4 to see which pages people find and use. It records page views, how far you scroll, video plays and file downloads, plus where the visit came from, device type, approximate location and a generic event when an embedded checker is first used, without its hardware selections or verdict. Standalone private tools do not load Google Analytics. The page address and referring address we send never include the part after a ? or #, so your selections never reach Google; the one exception is a short utm_source label that another site, such as chatgpt.com, adds to its links to say where you came from; clicks on outside links are not recorded, because some depend on the card you picked. Google signals and ad personalization are turned off, and we show no ads.

In the European Economic Area, the UK and Switzerland, and whenever your device's time zone is in Europe, Google Analytics sets no cookies until you choose Accept (the banner appears when your time zone is in Europe). Until then, and if you decline, Google may still receive measurement requests without cookies or a stored identifier; like any web request they carry your IP address and browser details, and Google uses them for basic measurement and modeling (Consent Mode). Elsewhere, Google Analytics sets the first-party cookies _ga and _ga_*, which expire 13 months after they are set; later visits do not extend them. Your choice is kept in your browser's local storage; declining deletes those cookies in every open tab. Google processes this data under the Google Privacy Policy.

Microsoft Clarity

Microsoft Clarity records how pages are used: mouse movements, clicks, scrolling and the page as it looked, which we see as heatmaps and replayed sessions, plus device type, browser and approximate location. Since 13 October 2026, Clarity may record the outer home page and checker guides. Every checker runs in an iframe on our separate origin, tools.localrigcheck.com, where Clarity, Google Analytics and the Cloudflare analytics beacon do not load. The outer page cannot read the checker's contents, storage, selections, dynamic links or results. The frame has a fixed height; no result, resize or configuration messages are sent to the outer page. It receives only fixed ready/first-use signals. Query-bearing visits, shared selection links and visits whose referring address contains query parameters are not recorded by Clarity. Clarity still records ordinary browser/device information about the visitor, as listed above; it never receives the hardware entered into the checker. Clarity loads only where Google Analytics may set cookies: outside Europe unless you choose Decline, and in Europe only after you choose Accept on the same banner. Where Clarity runs, it sets the first-party cookies _clck (up to a year) and _clsk (one day) to group page views into a visit, and drops them while Google's consent state for analytics is denied (in the European Economic Area, the UK and Switzerland that is until you accept). Declining stops Clarity and deletes those cookies. Microsoft processes this data under the Microsoft Privacy Statement.

Our own daily counter records public page views and private tool loads, then the first occurrence of each operation category per tool load: checker use, changing a control or preset, submitting, attempting to copy code or share a result, opening a full checker and invalid input. The Qwen-Image, Kolibri, MiniMax H3, Nemotron 3.5, FLUX 3 Image Klein, GLM-5.3-Flash and DeepSeek V4.1 Flash tools also count first download/deployment-route clicks, and every model checker except the Mistral Large 4 planner counts the first click on its "Other ways to run it" links (shown after a "does not fit" or "with caveats" answer). It stores only the UTC date, public page path, event category and count. It stores no field names, values, link targets, hardware verdicts, cookies, IP addresses or visitor/session identifiers. Requests carry no cookies or referring address. A per-IP edge rate limit uses the request IP transiently to reduce junk; the IP is not written to the counter database. Historical Strata counts from before this change still include its overall verdict. These daily aggregates cannot reconstruct or identify an individual user's sequence of actions.

The same counter counts fetches by automated clients, so we know whether search engines and AI assistants read the site. When a request for a public page, or for robots.txt, llms.txt, the sitemap, the feed or a file under /data/, names a bot, crawler or scripting tool in its user agent (the software name every request carries), has no user agent, or is signed by a browsing agent (a Signature-Agent header), our server sorts it into one of four fixed categories: an AI assistant fetching a page for a person (such as ChatGPT-User or Claude-User), a search index, AI training, or another bot. It then adds one to that day's count for that page path and category. The user agent is read only in memory to pick the category and is not stored; neither is the IP address, the part of the address after a ? or #, any cookie or any identifier. Ordinary browsers are not counted this way, fetches of the private tool origin are not counted this way either, and each category stops counting after a fixed number per day.

The RunPod and Vast.ai links in the checkers' "Other ways to run it" block are referral links: the address carries RigCheck's referral code, and nothing about your selections or your visit. What RunPod or Vast.ai then records is covered by their own privacy policies; their referral programs may tell RigCheck that a referred account signed up or paid, so that RigCheck can be credited.

Email alerts

If you sign up for email alerts, we store your email address, the kinds of hardware you picked, a random token used in your confirmation and unsubscribe links, whether you have confirmed, the sign-up date, when an unconfirmed sign-up expires, the day we last sent you a confirmation email, which alerts and on which day we last sent you, and an internal id of the last digest send, used to undo it if the email service refuses it. We use them only to send the alerts: one confirmation email, then at most one email a day, only when there are new model checkers you have not been told about yet (it can come a day after they go live). We never share or sell the list. The alert pages load no Google Analytics or Microsoft Clarity.

Nothing but the confirmation email is sent until you click its link, and at most one confirmation email a day; the link stops working 48 hours after the first request, asking again does not extend that, and the address is deleted within the following hour. Unsubscribing, with the link in any of our emails or the one-click button your mail app may show, deletes your row from our list, not just marks it. The links put the token after the #, which browsers do not send to servers; the page then removes it from the address bar and keeps it only in that browser tab's history, so a reload still works, until you press the button. The one-click button sends it to our server only to delete the row. The sign-up form's request IP is used transiently for the same per-IP rate limit as above and is not stored. Our daily counter adds the number of alert emails sent each day, without addresses. It also counts, per day, how many people reached the sign-up page from each kind of link, signed up and confirmed, by where they started (for example a graphics card page or a checker) and nothing else: this is not stored with your address, and the confirmation link carries the starting point after the # only so that the confirmation can be counted. The sign-up form inside a checker sends only what you see in it (your address, the hardware you left selected and where you started), and only when you press its button; the checker's other choices are not sent. Each email is sent at most once: if the email service does not answer, we do not try again later, so an email can occasionally be lost but never arrives twice.

The emails are sent through Resend, which receives your address and the email's content to deliver it and keeps sent emails and their delivery logs for 30 days (Resend security), after which they are deleted. We have not turned on Resend's open or click tracking, so the emails contain no tracking pixel and their links are not rewritten. To have your address deleted without the link, email us.

Measurement reports

If you send a report through the measurement form, we store what you entered (whether it ran, the model and file, the hardware, engine, context length, the pasted output or log, your notes and the optional name to credit), the UTC day, a random id, which version of the CC BY 4.0 agreement you ticked and whether the report is still to be checked or published. We do not ask for an email address and store no IP address, cookie or browser details with it; the request IP is used transiently for the same per-IP rate limit as above. Our daily counter adds the number of reports received each day, without their contents. The form page loads no Google Analytics or Microsoft Clarity, and the preview of what we read from your output is made in your browser.

Reports are read only by us until they are checked. A checked report is published on the model's page under CC BY 4.0, and in our open data files where the page's file carries measurements: the hardware, engine, settings and numbers, with the name you gave or anonymously; the full log is kept as the record behind the number and quoted only as far as needed to show it. Reports the log does not support, duplicates and spam are deleted when they are checked. Remove anything personal from a log before sending it; if something slipped through, or you want a report deleted, email us with the day you sent it and the model, and we delete it (a number already published stays in the open data copies others may have made, as the licence allows).

Questions: hello@localrigcheck.com.