streamlit-analytics2

Privacy-first usage analytics for Streamlit apps. One with block gives you pageviews, visitors, widget clicks, custom events, a built-in dashboard and SQL over your own data. No JavaScript, no cookies, no IP addresses stored.

Use it

1. Install:

   pip install streamlit-analytics2
   

2. Wrap your app:

   import streamlit as st
   import streamlit_analytics2 as sa2
   
   with sa2.track():
       st.write("Hello")
       st.button("Click me")
   

3. Open your app with ?analytics=on on the URL:

   http://localhost:8501/?analytics=on
   

That is the whole integration. Everything below is optional.

Keep the numbers across restarts

with sa2.track(save_to_json="analytics.json"):   # counters + analytics.events.jsonl
    ...

with sa2.track(events_path="analytics.db"):      # SQLite, unlocks the query tab
    ...

Protect the dashboard

with sa2.track(unsafe_password=st.secrets["analytics_password"]):
    ...

The password is plain text inside the app, so keep it in st.secrets or an environment variable and pick something you would not reuse.

What the dashboard shows

• Views, visits, visitors, bounce rate, average visit time, active now.
• Views per page over time, hourly for today, with range and page filters.
• Pages, widgets, browsers, OS, devices, languages, regions, UTM sources and campaigns, custom events.
• A weekday-by-hour traffic-load heatmap and the busiest hour.
• Recent visits.
• Raw data query: read-only SQL over the SQLite log, example queries, CSV download and a chart picker. Needs events_path on a .db file and a password.

Your own runs with ?analytics=on open are not counted.

Track your own events

if st.button("Generate report"):
    sa2.event("report generated", rows=len(df))

Privacy

Stored per visit: a hash of address and browser that changes every day, browser, OS and device family, language, timezone, theme, UTM tags, pages. Stored per interaction: widget type, label, key, page and the chosen option.

Never stored: IP addresses, raw User-Agent strings, query strings, or anything typed into text fields. Set store_values=True if you do want typed text.

Options

start_tracking() and stop_tracking() take the same arguments if you prefer them to the with block. Every event is a plain record (ts, kind, session, visitor, page, name, widget_type, key, value, props), so the log is easy to feed into pandas, DuckDB or an LLM.

Multipage apps

Call sa2.track() on every page. Views and widgets are recorded per page in the event log and on the dashboard. The legacy counters in streamlit_analytics2.data are shared across pages, as before.

How it works

Streamlit already knows which widget a user changed on each rerun. This library reads that from one place instead of wrapping every st.* function, so widgets inside columns, forms, expanders, tabs, dialogs and the sidebar are all seen, and a widget rendering with its default is never counted as a click.

Upgrading from 0.10

Nothing to change in your code. The numbers will be lower because a widget rendering with its default no longer counts as an interaction, typed text is no longer stored unless you ask, and the dashboard reset now needs a password. Details in CHANGELOG.md.

Contributing

Issues and pull requests are welcome. See CONTRIBUTING.md. Development: uv sync --all-extras && uv run pytest. A dev app with every widget type lives in examples/dev/; examples/dev/seed.py fills a dev log with made-up traffic so the dashboard has something to show.

License

MIT. See LICENSE.