The story
Why Pyxis exists
The introduction post explains why it was built and what it never stores.
Open-source product analytics
Product analytics without cookies, IP addresses or personal data.
Pyxis shows which features get used, where a sign-up loses people and which requests fail. It stores no cookie, no IP address, no user agent and nothing personal, and each of those rules lives in code, covered by tests.
No sign-up. Invented data. It opens straight on the dashboard.
Demo StoreLast 30 days
Overview
- Visits
- 2,793
- +5.6% (+147)
- Identified users
- 189
- +9.2% (+16)
- Conversions
- 124
- −0.8% (−1)
- Write error rate
- 2.2%
- +0.6 pt
This periodPrevious period
Top pages
/dashboard2,030/1,409/orders1,211
Privacy
What never gets stored
Five things an ordinary analytics tag would send, and what is left of each one. The browser cuts first; the API checks again before anything reaches the database.
Page addressIn the browser
An ordinary tag would send:https://shop.example/orders/48213?email=ana@example.com&utm_source=news
What is left:/orders/:id · utm_source=news
IP addressAt the API: country and state, then the address is dropped. The city is never read.
An ordinary tag would send:203.0.113.7
What is left:BR · BR-BA
BrowserAt the API: classified, then dropped
An ordinary tag would send:Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/141.0 Safari/537.36
What is left:desktop · Chrome · Windows
Event propertyAt the API: anything shaped like an email, a phone number or a tax id
An ordinary tag would send:"coupon": "ana@example.com"
What is left:dropped
VisitIn the browser: no cookie, nothing that links two tabs
An ordinary tag would send:cookie: visitor_id=8f2c41d0…; Max-Age=34560000
What is left:sessionStorage · ends with the tab
A browser that sends “do not track” (Global Privacy Control or Do Not Track) switches the tracker off before anything is queued. Events are deleted after 13 months.
Dashboard
What the dashboard answers
Eight screens, each built around one question. Every card opens that screen in the live demo.
Overview
How many visits, conversions and failed writes this period, against the one before?
Open in the demoAcquisition
Which channels, sources and campaigns bring visits that convert?
Open in the demoDevices
Which devices, browsers, countries and states do visits come from?
Open in the demoFeatures
Which events and screens get used, and how does one property break down?
Open in the demoFunnel
Where does a sign-up lose people, and how long does each step take?
Open in the demoRequests
Which routes of your API fail, how often and how slowly?
Open in the demoVisits
Which visits did one thing, filtered by page, event, channel, device or failure?
Open in the demoTimeline
What did one visit, or one signed-in person, do, in order?
Open in the demo
Install
What a site adds
One package and two lines. Page views are recorded on their own; a named event is one call.
npm install pyxis-analyticsimport { init, track } from 'pyxis-analytics';
init({ key: 'pyxis_pk_…', endpoint: 'https://api.pyxis-analytics.dev' });
track('signup_completed', { method: 'email_code' });Keys are issued by whoever runs the API, and there is no public sign-up today. All three repositories are MIT-licensed, the AWS infrastructure (Terraform) included.
In production
Measuring real sites since October 2026
Neither site needs a cookie to be measured.
Engineering
Built in public
- 100% of statements, branches, functions and lines covered in all three repositories. CI fails below that.
- Integration tests against a real Postgres, and end-to-end tests in a real browser with axe accessibility checks.
- The tracker is published to npm from GitHub Actions, with provenance.
- Decisions written down as architecture decision records in each repository.
- A Storybook where every story is also an accessibility test.
Architecture
How the pieces fit
- Your sitepyxis-analytics in the browser
- CloudFrontIn front of the API
- APINestJS on AWS Lambda, in São Paulo
- PostgresNeon, reached by a role that cannot change the schema
- DashboardNext.js on Vercel, reads the API
- Daily jobDeletes events after 13 months
- Weekly digestAn e-mail every Monday
Compared
Pyxis and a typical cookie-based tag
| Aspect | Pyxis | Typical cookie-based tag |
|---|---|---|
| Cookie on your visitors | None | An id cookie, usually kept for months |
| IP address | Read for country and state, never stored | Depends on the tool and its settings |
| Links the visits of an anonymous visitor | Never: a visit is one tab | Yes, through the cookie |
| Third parties | None | Depends on the tool; sometimes an ad network |
| Source code | MIT, on GitHub | Varies |
“Typical” describes a common setup, not any one product.
FAQ
Questions people ask
Does Pyxis set cookies on my visitors?
No. A visit’s id lives in the tab’s sessionStorage and ends when the tab closes. Cookies exist only in the Pyxis dashboard itself, for the people who use it, never on the sites it measures.
Does it store IP addresses?
No. CloudFront tells the API the country and the state of each request; the address itself is never written anywhere, and the city is never read.
Can it tell that two visits came from the same person?
Only for people signed in to your site, through your own internal id, never an email. Anonymous visits are never linked: not across tabs, not across days.
Can a visitor opt out?
Global Privacy Control and Do Not Track switch the tracker off, and a site can offer its own switch with optOut() and trackingStatus().
How long is data kept?
Thirteen months, then a daily job deletes it. A person’s events can be exported or erased from your backend at any time.
Can I use it on my site?
The code is MIT-licensed on GitHub. Today keys are issued by whoever runs the API; there is no public sign-up.
Read the post