Under construction · AGPL-3.0
Open-source product analytics for Next.js: page views, funnels, retention and journeys, in one app on your own Vercel and Postgres. No cookies before a site identifies someone, and 1.5 KB in the browser.
A hosted version isn’t available. The npm package isn’t published yet: today you run Trailic from the repo.
How it works
One line in your layout counts page views and route changes. track and identify do the rest, when you call them.
The collector checks the key, the domain and the limits, and an event never touches the database. Batches wait in Vercel Queues, or a folder on a server of yours.
A job writes them in batches, so a serverless database can sleep. A failed write loses nothing, and a repeated one counts nothing twice.
On screens, or from an assistant through MCP. Every page says how old its data is, because “now” would be a lie.
What it shows
Each of these is a screen in the app and, except journeys for now, a tool an assistant can call.
Two to eight steps, each a page or an event, and how long people have from the first. It shows how many reached each step, in order, and how many stopped there.
Someone anonymous can only convert within a day, because their hash rotates. Identify them and a funnel can run for weeks.
Cohorts by the Monday of the week someone was first identified, and what share came back each week after, by any activity or by the event you choose.
Identified people only. An anonymous visitor is someone else tomorrow, on purpose.
From any page: each visitor’s first time on it that day, then the next three pages within half an hour, most common first.
Journeys follow each visitor who reached the page, so they’re the one screen that slows down as a page gets busier.
From an assistant
The MCP server is at /api/mcp. Add that address to Claude, ChatGPT or any MCP client as a remote connector: the client registers itself, you sign in and approve it, and each tool runs as you, with the same rules as the screens.
Changing or deleting a site, erasing someone and managing members stay on the screens: no assistant can do them.
These only read
These add what you asked for
Privacy
Anonymous visitors are counted by a hash that lasts one day on one site, computed on the server. It costs some accuracy, and that’s the trade.
utm_* tags, with emails and long tokens taken outidentifyWhether your site must ask for consent depends on where it runs and what it does: that decision is yours. This is general information, not legal advice.
How the privacy approach worksWhere it stands
Don’t rely on it as your only analytics yet.
| Overview, biggest site, 90 days | 70 ms |
| Funnel, three steps, 90 days | 130 ms |
| Retention, 13 weeks | 150 ms |
| A flush of 4,500 batches | 4 s |
5 million synthetic events over 120 days on 30 sites, on Postgres 17 limited to one CPU and 4 GB. Synthetic events, not real traffic.
Questions
No, a hosted version isn’t available. Today you run Trailic yourself, from the repo.
Not yet: the package isn’t published. The library lives in the repo, in packages/trailic.
Trailic has no licence fee, no seats and no credits. You pay whoever runs your app and your database, at their prices.
Trailic sets no cookies and keeps nothing in the browser until identify. Whether your site must ask for consent depends on where it runs and what it does, and that decision is yours. This is general information, not legal advice.
The library is built for the Next.js App Router. The collector is a plain HTTP endpoint that any browser or server can post to, but the component that counts page views is for React and Next’s router.
No. It’s a Next.js app and a Postgres database, so anywhere that runs Node works. Away from Vercel, you point the buffer at a folder on a disk that lasts and call the two jobs on a schedule.
A visitor counts once per site per UTC day, known bots are left out, and everyone behind one address with the same browser is one visitor, which is a known undercount.
From the blog
The code is on GitHub under the AGPL-3.0. Questions and bug reports go to its issues.