How many people sign in to Quilcast a day: 49 to 75 accounts, counted without tracking anyone
Quilcast counts how many accounts used it each day, and the number has been between 49 and 75 on every full day since 7 November 2026: 62.6 on average over the 12 full days from 7 to 18 November. It is a count of account
Quilcast counts how many accounts used it each day, and the number has been between 49 and 75 on every full day since 7 November 2026: 62.6 on average over the 12 full days from 7 to 18 November. It is a count of accounts whose browser sent Quilcast's server a request signed with their Farcaster key, once per account per UTC day, and the server keeps the number and nothing that says who they were. That makes it an honest "how many people opened the app" figure, but it is not a count of new users, and this post is about exactly what it does and does not measure.
I read the counter's source in the Quilcast repository (server/stats.mjs, version 0.100.0, commit 15c2666) and the live numbers at quilcast.com/api/stats on 2026-11-19. I also ran the file's own tests, which pass (8 of 8, Node 22.23.5).
How many people use Quilcast a day?
These are the signed-in accounts per UTC day, as GET /api/stats returned them on 2026-11-19:
| Day | Accounts |
|---|---|
| Fri 6 Nov | 35 |
| Sat 7 Nov | 54 |
| Sun 8 Nov | 49 |
| Mon 9 Nov | 55 |
| Tue 10 Nov | 64 |
| Wed 11 Nov | 68 |
| Thu 12 Nov | 58 |
| Fri 13 Nov | 68 |
| Sat 14 Nov | 67 |
| Sun 15 Nov | 59 |
| Mon 16 Nov | 64 |
| Tue 17 Nov | 75 |
| Wed 18 Nov | 70 |
Two of these days need a note. The counter shipped in Quilcast 0.72.0 on 6 November, committed at 15:38 UTC, so the 35 on that Friday is only the part of the day after it went live. And the count for the current day grows until midnight UTC: the endpoint showed 56 for 19 November when I fetched it a little after 10:00 UTC, so I left that day out of the average. The same thing explains why I first read 47 for 16 November on the 16th, while the finished day shows 64.
Over the 12 full days from 7 to 18 November, the average is 62.6 accounts a day. Weekends run a little lower than weekdays here (57.3 on the two Saturdays and two Sundays against 65.3 on the eight weekdays), though with four weekend days I would not call that a pattern yet.
What counts as a signed-in account?
The counter does not watch a login screen. It counts an account when the server receives a request that account's Farcaster key has signed. In server/looks.mjs, a function called keySigned checks that the request carries an ed25519 signature over <what>:<fid>:<timestamp>:<body> made in the last five minutes, and that the signing key is one the Farcaster node lists for that account. Only after both checks pass does it call seen(fid), which adds the account to today's count.
Which requests are signed? Reading the saved mutes and blocks, the synced theme settings, drafts, pins, notification settings and profile looks all are. In the app, syncMutes() runs when an opened account is signed in, and it reads the account's mutes from our server with a signed request. So an account that opens Quilcast in a browser holding its key shows up in that day's count without doing anything else.
That gives three limits worth stating plainly:
- It is accounts, not people or visits. One account that opens Quilcast on a phone and a laptop on the same day counts once. Someone with two Farcaster accounts counts twice.
- It is not new sign-ups. An account that has used Quilcast every day for a month is in each of those days' counts. The number tells you how many accounts were active, not how many arrived.
- Browsing without signing in is not counted. Quilcast works for reading without an account, and those visitors are invisible to this number. So are signed-in accounts that never made a request signed by their key that day.
This is also why the number sits so far from the other figure I can see for Quilcast. Google Search Console showed 5,985 impressions in the 28 days to 18 November, but almost all of them are for profile names and unrelated strings, so impressions say little about how many people use the app. The sign-in count is the one signal that comes from people actually using it.
How the counter avoids tracking anyone
The file's own header says "Counts, never people", and the code shows how. When an account's request is counted, the server computes HMAC-SHA256 over seen:<fid> with a random 16-byte salt that was generated for that UTC day, and keeps the first 32 hex characters in a set. If the same hash shows up again that day, it is skipped, which is how an account counts once. The set and the salt are written to today.json so that a restart in the middle of the day does not count the same accounts again.
At the next UTC day the file is replaced, salt and hashes together. After that, nothing in what is kept can be traced back to an account: counts.json holds only { day: number } pairs, and the endpoint serves the last 90 days of them. Because the salt is different every day and is deleted afterwards, the hashes of two days cannot be matched to find out whether the same account came back. This is a deliberate trade: we get the number of accounts per day and give up the ability to ask how many came back.
The server also defends the numbers from its own mistakes. If today.json is lost, the day's count is kept at least at the larger of what was already counted and what the file had; a file that will not parse is moved aside rather than overwritten (the counts) or deleted (today's hashes, so nobody outlives their day). The endpoint is public and cached for 60 seconds (cache-control: public, max-age=60, which I confirmed with curl -I), and each IP is limited to 60 requests per window.
What about the campaign counts?
The same file also counts, per campaign link, how many accounts signed in for the first time in a browser that arrived through that link. The campaigns array in the endpoint was empty on 2026-11-19, because no tagged links are running, so there is nothing to report from it yet. The design keeps the same rule: which accounts joined through a campaign today is held in memory only and is never written next to an account.
Why we count this and nothing more
Quilcast has no analytics script. The goal is a measure of "are people using this" that is not a tracking system, and for a small Farcaster client a count of active signed-in accounts is the figure that answers it. If you want the same thing in your own app, the pattern is short: count at the point where a request is already authenticated, hash the identifier with a salt that dies at the end of the day, and publish only the totals. It took one file of about 170 lines (including comments) here, with tests.
If you use Farcaster and want to try Quilcast, it is a client for Hypersnap that runs in your browser and on iPhone and Android. The count above includes whoever opens it today.
FAQ
Does Quilcast track me? The server keeps a daily total of accounts and, for the length of the day, a salted hash of each account that used a signed request, which is deleted at the end of the UTC day. It stores no analytics script, cookie or IP address for this count.
Where can I see the numbers? At quilcast.com/api/stats, which is public, as JSON, for the last 90 days.
Is the source open? Yes, the counter is server/stats.mjs in backmeupplz/quilcast.
Numbers read from quilcast.com/api/stats on 2026-11-19 around 10:00 UTC; source read at Quilcast 0.100.0 (commit 15c2666); stats tests run on Node 22.23.5.
Originally published by Dev.to WebDev. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.