whatmodel.sh / privacy
Privacy policy
Last updated 8 October 2026
whatmodel.sh is operated by Merlinsbeard Pty Ltd (“we”, “us”), which is responsible for the personal information described here. This policy explains what information we collect when you use the website, its text and JSON endpoints (including from the command line), and who we share it with. There are no accounts, we don't use advertising trackers, and we don't sell or share personal information for advertising. Questions: [email protected].
1. Summary
- Visiting in a browser: the page loads DataFast analytics, which records the page you view and basic device and location information. No cookies.
- AI agents (like ChatGPT or Claude fetching a page for someone): our server reports the request to DataFast.
- Command line (curl and similar): not sent to analytics. Only our hosting provider's (Vercel) standard request logs.
- Hardware details you choose to send (memory, graphics card): used to build your list, not stored in a database, but they're part of the page address, so they appear in the analytics and logs above.
- Benchmark submissions (optional): if you run our bench script and choose to share, you post the report you reviewed as a GitHub issue under your own account. We review it and may publish the measurements. Reports contain no names, hostnames or file paths.
2. What we collect and why
Hardware details. To tailor the list you can send your memory size, graphics memory and, optionally, your chip or graphics card name, either by clicking the picker or by running one of the commands shown on the site. These are sent as part of the web address (for example ?ram=32&gpu=apple). We use them only to build the response. The commands shown only read these values on your computer and send them; they don't install or run anything else. The “your machine” hint (your operating system and graphics vendor) is worked out inside your browser and isn't sent to us.
Analytics for browser visits. When you open a page in a browser, a script from our analytics provider, DataFast, runs in cookieless mode and sends DataFast: the address of the page you view (including its parameters, such as the hardware you picked), the referring page, your browser, operating system and device type, and your approximate location (country, region, city), which DataFast works out from your IP address. The script and what it sends go through our domain, and our host passes your IP address on to DataFast for this. It doesn't set cookies. To count visits it keeps a short-lived identifier in your browser tab's session storage (cleared when you close the tab) and uses a hashed identifier derived from your IP address, browser user-agent and our domain, with a salt that changes about every 24 hours, so you aren't recognised from one day to the next. We use this to understand which pages and features are used, so we can improve the site.
AI agents and crawlers. When a request comes from a known AI agent or crawler (identified by its user-agent, for example ChatGPT-User, Claude-User, PerplexityBot or Googlebot), our server sends DataFast the address requested, the agent's user-agent, its IP address and the response status, so we can see how AI tools use the site. Requests that don't identify as a known AI agent aren't reported this way.
Command-line requests. Requests from tools like curl don't run the analytics script and aren't reported to DataFast. They're only recorded in hosting logs (below).
Hosting logs. Like any website, our hosting provider, Vercel, records standard request logs (IP address, time, address requested including its parameters, user-agent, response status) to operate and secure the service.
Benchmark script and submissions. The bench script probes local model servers and runs only the tests you select and confirm. API and agent tests contact your chosen provider, which receives the test request and normal connection information. API keys are kept in memory; agents use their existing login. The public report excludes keys, endpoint URLs, hostnames, paths, generated text and raw logs. It includes your chosen public model labels, runner, OS family, architecture, RAM and numeric measurements. Review your model labels for private information too. Results are saved locally. If you choose to share, the script opens a prefilled GitHub issue: opening the draft sends its contents to GitHub, and clicking Submit publishes it under your GitHub account. The script never submits automatically. Accepted measurements may be added to the public catalogue; submitted measurements are released under CC0 1.0.
Email. If you email us, we use your address and message to reply and keep a record of the conversation.
We don't use this information to make automated decisions about you, and we don't combine it with other data to identify you.
3. Legal basis
Where laws such as the GDPR or UK GDPR apply, we process analytics, AI-agent reports and hosting logs based on our legitimate interests in running, securing and improving a free public website, and we've chosen a cookieless, daily-rotating setup to keep that processing minimal. We process emails to respond to you, and benchmark submissions because you choose to send them. You can object at any time (see section 7).
4. Who we share it with
- DataFast, our analytics provider, which processes the analytics and AI-agent data on our behalf. See its privacy policy and data processing agreement.
- Vercel, our hosting provider, which serves the site and keeps the request logs. See its privacy policy.
- Our email provider, if you email us.
- Authorities, if required by law.
These providers may process data outside your country, including in the United States or the European Union. Where required, transfers rely on appropriate safeguards such as standard contractual clauses in the providers' terms. Sponsored placements are plain links; sponsors receive no information about you from us. Links to model providers, benchmark sites and other third parties are governed by their own policies. Fonts and icons are served from our own domain.
5. How long it's kept
We only keep information for as long as necessary for the purposes above. Hardware details aren't stored beyond the analytics and logs described here. Analytics data is kept in our DataFast account until we delete it or close the account, and is only linkable to a visitor within a single day. Hosting logs are kept for Vercel's standard retention period. Benchmark reports stay in the GitHub issue you posted, which you can edit or delete; accepted measurements are published indefinitely. Emails are kept as long as needed to deal with your message.
6. Children
The site isn't directed at children and we don't knowingly collect information from them.
7. Your choices and rights
You can opt out of analytics in this browser here. This stores a single opt-out flag in your browser, which DataFast checks before sending anything:
checking…
Pre-rendered model lists remain readable with JavaScript blocked; interactive filters require JavaScript. The analytics script can be blocked independently, and analytics is disabled in local previews. Depending on where you live, you may have rights to access, correct or delete personal information about you, to restrict or object to its processing, and to data portability. Because we don't keep profiles we may not be able to link any data to you, but email us and we'll do what we can, including for data DataFast holds on our behalf. If you're in the EU or UK you can also complain to your local data protection authority, and in Australia to the Office of the Australian Information Commissioner (oaic.gov.au). California residents: we don't sell or share personal information as those terms are defined in California law.
8. Changes
We'll update this page if what we collect changes, and change the date above. Significant changes will be noted on the site.