feat: first-party analytics tracker (no dashboard)

[dimslaev]

What

Self-hosted, third-party-free pageview + event tracking. Events are written to our own Postgres (analytics_event table). No dashboard — you query the table directly in SQL.

Why

Wanted analytics without a third party (no PostHog/GA) and without standing up a separate service (no ClickHouse/Matomo container). This reuses the DB we already run.

Backend

• models_agentique.py — new AnalyticsEvent table (nullable user_id FK, JSON props, indexed event/path/visitor_id/created_at) + AnalyticsEventCreate request model.
• api/routes/analytics.py — POST /analytics/collect. Public; attaches the user via CurrentUserOptional when a valid bearer token is present, records anonymously otherwise. Truncates strings to 2048, returns 204.
• Alembic migration (a7b8c9d0e1f2, head), conftest.py teardown, and test_analytics.py (anonymous/authed, default event, custom props, truncation).

Frontend

• lib/analytics.ts — trackPageview / trackEvent POST to the backend via fetch(keepalive). Persists an anonymous visitor_id in localStorage, works logged-out, and swallows all errors so analytics can never break the app.
• main.tsx — pageview fired from router.subscribe("onResolved", ...).

No double-sends on re-render

Tracking lives in the router subscription at module scope, not in a component render or useEffect — so React re-renders and StrictMode's dev double-invoke can't trigger it. A 500ms same-path guard covers edge cases (duplicate onResolved, dev HMR re-registering the subscription).

Notes

• No generated-client change — the tracker uses a plain fetch, so no sdk.gen.ts/types.gen.ts regen needed.
• The endpoint is public and unauthenticated (by design, for anonymous pageviews). Add rate limiting if abuse becomes a concern.

Querying examples

-- top pages last 7 days
select path, count(*) from analytics_event
where event = 'pageview' and created_at > now() - interval '7 days'
group by path order by 2 desc;

-- unique visitors per day
select date_trunc('day', created_at) d, count(distinct visitor_id)
from analytics_event group by d order by d;

🤖 Generated with Claude Code

https://claude.ai/code/session_01VjK4vCPYkTE18K6nMk3VDy

---

Generated by Claude Code