Analytics

Self-hosted website analytics, single-operator. One axum binary with sqlx + SQLite, minijinja templates, and a Vite + Bootstrap frontend. Tracks page views, clicks, scrolls, sessions, and custom events; renders metric cards, time-series charts, a world map with admin-1 drill-down, and PDF/markdown reports for any date range.

Features

Embed a small collector script on any number of properties (tracked sites)
Dashboard with 16 metric/chart/list aggregations and user-defined custom cards
GeoIP enrichment (auto-downloads DB-IP City Lite) and UA parsing (auto-downloads ua-parser regexes)
Bot traffic routed to a separate table so human dashboards never have to filter it
Public/private toggle per property; signed-cookie auth for the operator
PDF and markdown export of any dashboard view (PDF rendered in-process via embedded Typst, no chromium)
Single-binary deploy via git push server master

System dependencies

Local dev needs all of these on your PATH:

Tool	Why	Version
`rustc` / `cargo`	Build the axum binary	2021 edition, current stable is fine (1.70+)
`bun`	Frontend deps + Vite + map builder	1.x
`make`	Run the dev/build targets	any
`pkg-config` + OpenSSL headers	Linked at build time on Linux	distro packages: `pkg-config`, `libssl-dev` (Debian/Ubuntu), `openssl-dev` (Alpine)

Install hints:

# Rust toolchain (recommended via rustup)
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh

# Bun
curl -fsSL https://bun.sh/install | bash

# System libs (Debian/Ubuntu)
sudo apt install -y build-essential pkg-config libssl-dev

# System libs (Alpine)
sudo apk add musl-dev pkgconfig openssl-dev

The Docker build (see Dockerfile) reproduces this on rust:alpine + alpine:3.23. If you only care about Docker, you do not need any of the above on the host.

Two more files are downloaded into data/ at runtime, no setup required:

data/db.mmdb: DB-IP City Lite (CC-BY-4.0). Refreshed if older than 30 days.
data/regexes.yaml: canonical ua-parser regexes (Apache-2.0).

If either download fails the server still boots: UA parsing falls back to a substring heuristic and GeoIP enrichment is skipped.

Quickstart

cp samplefiles/env.sample .env
# edit .env to set ANALYTICS_PASSWORD and BASE_URL
make

make (alias make run) installs frontend deps if needed, then runs Vite watch and cargo run concurrently on port 8000. Visit http://localhost:8000/login.

First boot creates a “Proprium” property and starts tracking the dashboard’s own usage.

Configuration

All config comes from .env (loaded via dotenvy). The full set:

Variable	Required	Purpose
`ANALYTICS_PASSWORD`	yes	Single operator password
`BASE_URL`	yes for prod	Used in absolute URLs (sitemap, og tags, embed snippet)
`PORT`	no (default `8000`)	HTTP listen port
`ANALYTICS_COOKIE_SECRET`	no	32+ bytes for signing the session cookie. Falls back to a SHA-512 of the password, so rotating the password invalidates sessions
`ANALYTICS_DATA_DIR`	no (default `./data`)	Where the SQLite db, mmdb, and regexes live. Production sets this to `/data`
`ANALYTICS_ROOT`	no	Override the project root (where `templates/`, `dist/`, `migrations/`, `static_maps/` are read from)

Make targets

Target	What it does
`make run` (default)	Vite watch + `cargo run` on port 8000
`make build`	Vite assets + topojson maps + release binary (`target/release/analytics`)
`make start`	Run the release binary (after `make build`)
`make maps`	Rebuild per-country topojson under `static_maps/` from Natural Earth
`make seed`	Create or refresh a “Seed Test” property with realistic fake events. Override with `SESSIONS=2000 DAYS=60`
`make migrate FROM=<path-to-django.sqlite3>`	One-shot import of an existing Django analytics database, preserving property UUIDs so embedded snippets keep working. Add `FORCE=1` to wipe first
`make pull`	rsync the production db + geoip from `git remote server` into `data/`
`make push`	`git push` to every configured remote
`make clean`	`cargo clean` plus removing `dist/`, `node_modules/`, the SQLite db, and the mmdb

There are no tests or linters configured.

Deploy

Production runs on Docker. The standard flow is git push server master to a remote whose post-receive hook runs docker compose up --build --detach. Sample files in samplefiles/:

Caddyfile.sample: reverse proxy with TLS
env.sample: the same .env shown above
post-receive.sample: the git hook

Data persists to /srv/data/analytics/ on the host (mounted into the container at /data).

Stack

Backend: axum 0.8, sqlx 0.8 against SQLite (WAL, synchronous=NORMAL, busy_timeout=5s), tower-cookies for signed sessions
Templates: minijinja 2 with a Jinja2-faithful HTML formatter
Frontend: Vite 6, Bootstrap 5 SCSS, Chart.js, d3-geo + topojson, monaspace argon font (self-hosted via @fontsource)
Enrichment: maxminddb (GeoIP), uaparser (UA)
PDF: embedded Typst (typst + typst-pdf + typst-kit), no chromium subprocess

See CLAUDE.md for the full architecture rundown, route table, and data model.