EmailKnow: an honest, open-source email tracker (0 users, and here's the whole thing)
Every email tracker I've used tells you the same comforting lie: "✓✓ Read."
Then Apple ships Mail Privacy Protection, which pre-loads the tracking pixel on Apple's servers before the human ever opens the message — and suddenly "Read" means nothing. The tools kept showing the green checkmark anyway. They'd rather look confident than be honest.
That bugged me enough to build the opposite.
I'm building EmailKnow: an open-source Gmail extension that tracks whether your email got read AND acts on it — but refuses to pretend it knows more than it does.
The one rule that shapes everything
No fake certainty. When someone opens your email, EmailKnow sorts the signal into four honest states, not one:
- 🔥 clicked a link — the strongest signal a real human was there
- ✓✓ human open — a real open we're confident about
- ⚠️ Apple/proxy pre-load — the pixel fired, but it might be a machine, not a person. We label it, we don't count it as "read."
- ⧗ unknown — we're not sure, and we say so
That ⚠️ state is the whole product in one pixel. Competitors collapse all of this into a single green check. I'd rather show you the truth and let you decide.
It's not just tracking — it's the follow-up
Tracking that just sits there is a vanity metric. The point of knowing "they read it but didn't reply" is to do something about it. So EmailKnow closes the loop: you set a rule, and it sends a follow-up in the same Gmail thread at the right moment — driven by the signal, not by a dumb timer. "Read but no reply" gets a different nudge than "never opened."
The privacy stance (this is not marketing)
Because I read a lot of these tools' privacy policies and they horrified me:
- Your email body and attachments are NEVER read or stored. The backend only ever touches header-level metadata — subject, thread ID — and events.
- IP addresses are stored only as a salted SHA-256 hash. Never plaintext. I can't tell you where you are; I can only tell two opens apart.
- Minimal Google scopes: send, and metadata (headers only). No scope that can read your message body even exists in the grant.
- It's open source, so you don't have to take my word for any of the above — you can read it.
Where I'm at (honestly)
- Users: 0
- Paying: 0
- Revenue: $0
Backend and the extension are built and tested; the whole thing runs inside Cloudflare's free tier, so my hosting bill is basically $0 too. No launch yet.
I'm posting this early because the honesty angle only works if I'm honest about the journey too.
So, a real question for anyone who sends email that matters — sales, outreach, job hunts: would you rather a tracker show you a confident "Read" that might be a lie, or an honest "we're not sure"? I genuinely want to know, because that's the line the whole product is built on.
— building EmailKnow in public, #1
I really like the “unknown” state. I think that's actually harder to build than it sounds.
There's an interesting distinction between a weak signal and a negative signal. An Apple/proxy pre-load isn't evidence that the person read the email, but it's also not evidence that they didn't. Treating both as the same would turn uncertainty into a false conclusion.
I'd be curious whether you eventually model these as confidence levels rather than fixed states — especially when multiple signals appear together. For example, a proxy open followed by a link click is very different from a proxy open with nothing else.
The principle of “don't manufacture certainty when the evidence isn't there” feels like the more interesting product philosophy here.
The weak vs negative distinction is exactly the trap I was trying to dodge. A proxy pre-load and a real "never opened" are worlds apart, but every tool I tried throws them in the same "not read" bucket. That's the false conclusion, right there.
On confidence levels: the states are already less fixed than they look. Signals stack, so a proxy open plus a later click gets bumped up, because the click is hard proof a human was there. What I won't do is slap a "73% read" number on it. The second you show a precise percentage, people trust it like gospel, and I'm back to faking certainty I don't have. I'd rather roll signals into a few honest buckets than invent a decimal.
Where'd you draw that line, though? Somewhere between "too vague to be useful" and "so precise people trust it blindly" there's an honest sweet spot, and I haven't nailed where it sits. Curious how you'd think about it.