Decks moved to HTML. I built the read tracking for it.

HTMLRadar is send-side analytics for HTML documents. You upload an HTML file or paste a URL you already host, create one tracked link per recipient, and send it. The dashboard gives you one row per viewer who opened it: which sections they read, how much active reading time they spent, how far they scrolled. You also get an email the first time someone opens it.

Every share carries its own controls - email gate, password, expiry date, an allow-list by domain or exact address, and revocation. One document, many shares. The password, expiry and allow-list can be changed after you've sent the link.

The mechanics are the interesting part, so I'll keep this to six sentences.

1. A Cloudflare Worker sits at /r/{slug}, checks that share's gates, fetches your HTML, and injects the tracker with HTMLRewriter on the way through, so the recipient sees your document, not a re-rendered copy of it.
2. The tracker itself is roughly 8 KB gzipped, and since the source is public you can read the whole thing before you decide whether to trust it.
3. Section detection runs a fallback chain over your markup: headings first (h1, h2, h3, or a selector you configure), then slide or page containers such as section, article and anything with slide or page in its class name, and finally paragraph buckets for plain prose.
4. Time on a section only starts counting once at least 50% of it has stayed visible for one continuous second, and the read signal fires after three qualified seconds, so scrolling fast past a slide never registers as having read it.
5. Both section dwell and session active time run a five-second idle watchdog: session time resets on keydown, scroll and touchstart; section dwell also resets on mousedown and wheel; mousemove is deliberately excluded, because a twitching cursor isn't reading.
6. Every proxied response carries a sandbox Content-Security-Policy without allow-same-origin, so the recipient's document runs in an opaque origin and can't reach application storage.

The whole list, so you can decide before you send anything. An email address when the recipient enters one at a gate, or a random browser ID when they don't; then the time of each open, referrer, the browser's user-agent string and the device type, browser and operating system read from it, coarse location, per-section dwell, scroll depth and active reading time, and, if the document has attachments, which ones were downloaded. There's no mouse tracking, no keystroke capture, no DOM snapshots, no session replay and no raw IP address. A recipient can start an opt-out with window.HTMLRadar.optOut() and confirm on the page that opens.

I send decks to investors and clients; I wouldn't send them a link that records more than that.

HTMLRadar ships an MCP server, so the agent that wrote the HTML can publish it as a tracked link, and you can ask that same agent the next morning whether anyone opened it. There are three tools over stdio - share_html, get_share_activity and whoami - and they work in Claude Code, Cursor and Codex CLI. In Claude Code a plugin layers a skill on top, so the agent knows when to offer a tracked link and when to stay quiet. I built it as a side path, but it's now the part I use most.

If you'd rather your documents didn't sit on somebody else's infrastructure, the whole stack self-hosts. You need a Cloudflare account (Workers, R2 and Pages), a Supabase project, and a domain on Cloudflare DNS. Cloudflare and Supabase both have free tiers. Resend, which sends the emails, is optional; without it the first-open trigger writes a skipped row and everything else carries on working. The schema is a folder of numbered, idempotent SQL migrations you paste into the Supabase editor in order. The self-hosting guide is in the repo, and the hosted version is deployed from the same repository.

The hosted version is free for two tracked links, then $15 a month or $150 a year for unlimited ones. Self-hosting has no licence fee and never will, because the code is AGPL.

Source, issues and the roadmap: https://github.com/htmlradar/htmlradar

Happy to hear your thoughts, particularly on the section-detection heuristics - the part I'm least confident about. Plain prose with no headings is hard to bucket well, and I suspect somebody reading this has a better idea than paragraph groups.

Cheers,
Abhinandan