Peek

Privacy-first web analytics for personal sites. A cookieless, sub-2KB JavaScript snippet beacons pageviews to a single-process Node server that hashes visitors daily, never stores raw IPs, sessionizes with a 30-minute idle cutoff, and serves a dark basic-auth dashboard with inline-SVG charts.

Built as a judged exercise: the dashboard was iterated against Plausible's own live public dashboard (plausible.io/plausible.io) until blind critics picked ours. Verdict logs in progress.json history / refs/.

Deploy (Render free)

1. Push this repo to GitHub, then on render.com: New → Blueprint → pick the repo.
2. In the service's Environment tab set DASH_PASS (and a long random SALT); SITE is the site name shown on the dashboard.
3. Deploy. The health check is /health.

Then drop the snippet into any page (GitHub Pages, etc.) — it beacons back to wherever it was loaded from:

<script defer data-site="your-site-name" src="https://YOUR-RENDER-URL/s.js"></script>

Real traffic in, dashboard out. Free-tier ceiling: the instance sleeps after ~15 min idle and the SQLite file lives on the instance disk — data survives sleep, but is lost on redeploy/recycle. For durable data, add Render's persistent disk (paid) or a volume host.

Honest ceiling

This is a personal-site tool, not Plausible. At serious traffic (say, tens of thousands of pageviews per day) the in-memory session map, single-process design, and geoip-lite lookup stop scaling, and you should self-host the real Plausible instead. Peek exists to give a personal site its own private analytics with zero external requests and a handful of files.

What gets stored

Per event, one row in SQLite: site, UTC day, timestamp, a daily-rotating visitor hash, session start, path, referrer host, screen class, country.

Never stored: raw IP addresses, full user agents, referrer URLs (only the host), query strings, cookies, fingerprints. The visitor hash is HMAC-SHA256(daySalt, ip|ua) where daySalt derives from SALT plus the calendar date, so the same visitor produces a different hash every day and cannot be correlated across days. Raw IP and UA are consumed by the hash and discarded.

Obvious bots (crawlers, curl, headless browsers, monitoring agents) are dropped by user-agent regex before anything is written, and get the same 204 response as real visitors so the filter is not detectable.

Quickstart

cp .env.example .env   # set DASH_USER / DASH_PASS / SALT
npm install
npm run seed           # optional: 30 days of demo traffic for the dashboard
npm start              # one process on :8000

Add the snippet to any page, pointing src at the Peek host:

<script defer src="https://analytics.example.com/s.js" data-site="demo"></script>

The snippet beacons path, referrer host, screen class (mobile/tablet/desktop) and site id via an image GET. No cookies, no localStorage, no fingerprinting. It fires once per page load (no SPA route tracking yet; add it if your site is a single-page app).

Dashboard: http://localhost:8000/dashboard (basic auth). Ranges: today / 7d / 30d.

Privacy properties

Property	How
No cookies	snippet uses a bare <img> GET
No raw IP stored	hashed with a per-day salt, then discarded
No cross-day correlation	hash changes daily by construction
No user agent stored	consumed by the hash, never persisted
Bounce rate	sessions with exactly one pageview / all sessions
Sessionization	server-side idle cutoff of 30 minutes (in-memory)

Known privacy ceilings, stated honestly: (1) the operator of this instance can correlate a visitor across their own sites (same IP+UA hashes identically on the same day) - fine for a personal site, not a shared service; (2) the visitor hash is daily, so 30-day unique counts are per-day uniques summed by SQLite's COUNT(DISTINCT) semantics only within a day; (3) trust proxy is set to loopback so a reverse proxy on the same host (nginx/caddy) is honored but remote X-Forwarded-For spoofing is rejected.

Checks

npm test

Verifies: snippet gzip size < 2 KB, daily hash rotation, bot filtering, 30-minute session cutoff, bounce math, no ip/ua columns in the schema, HTTP ingest (204, country resolution, silent bot drop), and dashboard auth.

Measured against the bar

Metric	Peek	Plausible (plausible.io)
Snippet, gzipped	see npm test output (target < 2 KB)	~1 KB
Dashboard TTFB (same connection, localhost vs public host)	see refs/measure.txt	measured

The dashboard is compared blind against plausible.io/plausible.io at desktop (1440x900) and mobile (390x844) in refs/. Screenshot ours with:

"/c/Program Files/Google/Chrome/Application/chrome.exe" --headless=new --disable-gpu \
  --user-data-dir=/tmp/p1 --window-size=1440,900 --virtual-time-budget=8000 \
  --screenshot="C:/Users/Spoidy/peek/refs/ours-desktop.png" \
  "http://admin:peek-admin-2026@localhost:8000/dashboard?range=7d"

Files

lib.js               core: db, hash, bot filter, sessions, queries, SVG chart
server.js            one process: snippet, ingest, dashboard, progress
public/snippet.js    the client snippet (minified)
views/dashboard.html dark dashboard template
views/progress.html  live build-progress page (served at /progress)
seed.js              backfill demo traffic through the real hash + session rules
test.mjs             self-check
progress.json        live state for /progress