DataMyna

A self-hostable, open-source product analytics tool. Drop in a script tag, get a dashboard.

Screenshots

Quick start

docker compose up

Open http://localhost:5173.

Seed the database with 120k demo events:

docker compose exec backend python seed.py
docker compose exec backend python -c "from aggregator import backfill; backfill(90)"

---

Embed the tracker

<script src="http://localhost:8000/tracker.js"></script>

<!-- Named event on any element -->
<button data-track="signup_click">Sign up</button>

<!-- Programmatic event -->
<script>
  datamyna.track("upgrade_click", { plan: "pro" });
  datamyna.identify("user_123"); // call after login
</script>

---

Architecture

tracker.js  →  POST /ingest  →  raw_events (append-only)
                                      │
                                 aggregator (hourly)
                                      │
                         ┌────────────┼────────────┐
                    dau_summary  top_pages   event_summary
                         │           │             │
                         └────────────┴────────────┘
                                      │
                             GET /dau, /top-pages …
                                      │
                              Vue 3 dashboard

User journey (drill-in):
  GET /sessions         → reads sessions summary table
  GET /sessions/:id/events → reads raw_events filtered by session_id (indexed)

---

Endpoints

Method	Path	Description
POST	/ingest	Receive event batches from tracker.js
GET	/dau?days=30	Daily unique users
GET	/wau?weeks=12	Weekly unique users
GET	/mau?months=6	Monthly unique users
GET	/top-pages?days=7	Top 10 pages by views
GET	/events/timeline?event_name=signup_click	Named event over time
GET	/stats/today	Today's KPI snapshot
GET	/sessions?days=7&user_id=u_001	Session list
GET	/sessions/:id/events	Full event sequence for one session

---

Scaling notes

The current setup uses pre-aggregated summary tables as a lightweight data mart. Raw events are append-only and never queried directly by the dashboard (except the per-session drill-in, which is an indexed lookup, not a scan).

At higher ingest volumes, natural next steps are:

• Redis queue between /ingest and the DB writer to absorb write spikes without blocking the API
• Parquet exports of raw_events for cheap long-term storage queryable with DuckDB
• TimescaleDB or ClickHouse for partitioned time series at very large scale, with compression and columnar storage
• HyperLogLog in the aggregator for exact WAU/MAU counts (the current approach sums daily uniques and may double-count cross-day users)