aydex.cloud

Privacy-first web analytics: cookieless by default, no recordings, and built to explain why your traffic changed — not just chart it.

Piece	Tech
web/	Next.js 14, Prisma (Postgres), NextAuth, Stripe, Recharts
tracker/	Go collector + the ~10 KB readable tracking script (script.js)
mcp/	MCP server so Claude can answer questions about your analytics
ClickHouse	Event store (analytics.pageviews, analytics.imported_daily)
Redis	Rate limits, monthly quota, live visitors

What's in it

• Parity basics — pageviews, visitors, sessions, bounce rate, time on page, scroll depth, entry/exit pages, countries, devices, UTM campaigns, goals, custom events, funnels, CSV export, public share links. Bots and referrer spam are filtered at collection time.
• Why traffic changed — rule-based explanations (dashboard, weekly email, API, MCP) computed from your numbers: which channel, referrer, page, country or device accounts for the change; new and lost sources; anomalies.
• Behavior signals without recordings — rage clicks, dead clicks, JS errors, outbound and download clicks. Only a count plus a CSS selector or scrubbed message is sent: no text, keystrokes or screenshots.
• AI-referrer tracking — visits from ChatGPT, Perplexity, Claude, Gemini, Copilot and others, as their own channel.
• Revenue by source — aydex.track('purchase', { revenue: 49, currency: 'USD' }), attributed to the visitor's first pageview source that day.
• Public REST API on every plan and an MCP server (mcp/README.md).
• Ad-block resistance — serve script and collector from your own domain (nginx / Caddy / Next.js recipes in the dashboard) with data-endpoint.
• GA4 history import — upload a GA4 daily CSV; it fills only dates before your first aydex pageview.
• Verifiable privacy — /privacy-proof hashes and statically checks the exact script served to visitors.

Tracking snippet

<script defer data-site="YOUR_TRACKING_ID" src="https://t.aydex.cloud/script.js"></script>

Optional attributes: data-storage="localStorage|sessionStorage" (+ data-key) for persistent encrypted visitor ids, data-endpoint="e" (first-party proxy), data-behavior="off", data-spa="off", data-respect-dnt="true"; add data-aydex-ignore to any element to exclude its clicks. Names starting with _ are reserved for internal signals.

Public API

curl -H "Authorization: Bearer ax_live_…" https://aydex.cloud/api/v1/sites
curl -H "Authorization: Bearer ax_live_…" "https://aydex.cloud/api/v1/sites/SITE_ID/insights?period=7d"

Reports: overview timeseries pages referrers channels ai-referrals countries devices browsers campaigns events engagement entry-exit behavior revenue funnels insights. Periods: today 7d 30d 90d. Limit: 120 requests/minute per key (HTTP 429 beyond that). GET /api/v1/stats/:siteId is kept for backward compatibility.

Definitions (so the numbers are honest)

• Session = one visitor id on one UTC day. Visitor ids are cookieless and rotate daily unless the site opts into data-storage, so a session never spans midnight.
• Time on page = visible-tab time per page load, averaged over page loads that reported it.
• Dead click is a heuristic (a button-like element that changed nothing for 800 ms); expect some false positives.
• Imported visitors are daily-user totals summed over days, not de-duplicated.

Run / deploy

See docker-compose.yml, deploy/ and scripts/gen-env.sh. Schema changes are additive (prisma db push creates the funnels table; the tracker creates analytics.imported_daily on start).

Develop

cd web && npm test && npm run typecheck && npm run lint
cd tracker && go vet ./... && go test ./...
cd mcp && npm test