Visitor analytics
Every page loads a script from Umami, an open-source analytics tool that this site runs itself at umami.agentmods.dev, on the same server as the catalogue. Nothing about your visit leaves that server. It records the page you opened, the site you arrived from, your browser, operating system, device class and country. It also receives the same moments this site counts for itself — a command copied, a repository opened, a watch requested, a rating given, how long a page stayed open, and that a search happened and how much it found (never the words typed) — so that one place can answer where a visit came from and what it did. It sets no cookies and does not store your IP address, which is used only to derive the country and then discarded. Because the site sends a strict-origin-when-cross-origin referrer policy, it sees the referring site, not the full referring page.
Requests that Cloudflare or the user agent identify as a crawler are counted separately as bot traffic and kept out of the visitor figures. There is no third party in this path: no analytics vendor receives anything about you.
The same Umami also records session replays and heatmaps: where on a page people click and scroll, and a replay of how the page changed while it was open. A replay is a recording of the page's own structure and of your pointer and scrolling, not a video and not a screenshot. Text you type into any field is masked before it leaves your browser, and so is text on the page at the "moderate" masking level Umami applies. Recordings stop after five minutes. They are stored on the same server as everything else, they are not shared, and they exist to show which parts of a page get used and which get ignored. If you would rather not be recorded, a browser extension that blocks umami.agentmods.dev stops both the replay and the page-view count, and the site works exactly the same without them.
What this site records itself
Two actions on a mod's page send a single message to this site: copying an install command, and clicking through to the source repository. The message carries the mod's numeric id and which of the two happened. Nothing else.
Every page also sends one message saying a page was viewed, and one as it closes saying roughly how long it was open. Neither carries an address, a path or an identifier of any kind. The time is rounded into four bands — under a second, under ten, under a minute, longer — before it is stored, and it is stored per country and per day, never per visit.
Alongside it, one more count: the two-letter country code your request arrived with, kept per day and per kind of event in its own table. Not with the mod, not with anything else, and never an address — the country is the finest this site records about where anyone is.
What you type into the search box is counted too, in the same shape: the query, the day, the country, and whether it came from the site, the API, the CLI or the MCP server, with a counter. It is lower-cased and trimmed first so that one thing is one row, and anything that looks like an email address, a link or a secret is dropped before it is written rather than stored and cleaned later. There is no row per search and nothing that joins one search to another or to a visit. The counts are public on the stats page; they exist to show what people look for and what the catalogue does not yet have.
It is stored in a table called events, which has four columns: the mod id, the kind of event, the date, and a counter. There is no row per person. There is no IP address, no user agent, no session, no identifier of any sort, and no way to reconstruct one visit from another. Requests that look like a crawler are dropped before they are counted.
It exists because how many people actually take a mod is ten points of the ranking on this site, and a catalogue that ranks mods without that signal ranks them on popularity of the repository alone.
The stack is different again: the mods you add to it live in your browser's local storage and are never sent anywhere. The page asks this site for the cards to display, and that request carries mod ids, not you.
The command-line tool
When npx agentmods add installs a mod it reports the mod's slug and the word "install" to the same counter. It sends no path, no project name, no machine or user identifier, and nothing about your configuration. If the report fails it is dropped silently, because telemetry is never worth failing an install over.
Three ways to turn it off, any one of which is enough:
npx agentmods add <slug> --no-telemetry
DO_NOT_TRACK=1 npx agentmods add <slug>
AGENTMODS_NO_TELEMETRY=1 npx agentmods add <slug>DO_NOT_TRACK is honoured because it is the standard way to say this once, for every tool on the machine.
Watching a mod by email
The watch form is guarded by Cloudflare Turnstile, which runs a short check in your browser to tell a person from a script before any mail is sent; Cloudflare already sits in front of this site, so that is not a new party. It exists so nobody can use this form to send confirmation mail to addresses that never asked.
This is the only place an address is stored, and it is stored only if you type one in. Asking to watch a mod records your email address, the date, and which mods you watch. Nothing more: no name, no IP address, no browser.
Your address is used for one thing, the digest telling you that a mod you watch changed. It is sent through Amazon Simple Email Service from [email protected], so that service handles the message in transit. Nothing is sold, and nothing is shared with anyone else.
Your IP address is looked at when you submit the form, to stop one machine from signing up thousands of addresses. It is held in memory for a few minutes and never written to the database.
Every mail carries two unsubscribe links. One stops a single mod. The other deletes your address and every watch attached to it, immediately and without asking anything further. Until you use one, the address is kept, because that is what makes the digests possible.
Repositories in the catalogue
The catalogue is built from public GitHub repositories. If you are the author of one, what is stored about it, and how to have it taken out, is on the crawler page.
Contact
agentmods is run by one person. Anything about this page, an address you want removed, or data you want a copy of: open an issue at github.com/iOSDevSK/agentmods/issues, or reply to any mail the site has sent you.