Litlyx

Last reviewed Sep 13, 2026 · 743 words

The feature that makes Litlyx interesting is the one that complicates its "privacy-first" label. Its built-in assistant lets you ask "which pages did visitors from Germany convert on last week" in plain English, and at last check that assistant works by sending your question and the relevant query results to OpenAI's API using a key you supply. The visit data itself stays in your MongoDB. Whether that arrangement counts as private is a question to settle before you install it, because the assistant is the reason to choose Litlyx over the alternatives.

The one-line script is real, and so is the MongoDB dependency

Integration is as advertised: one script tag with your project ID, and page views, referrers, countries and devices start arriving in a dashboard that looks more like a product analytics tool than a server log. There is a JavaScript API for custom events and a REST API for pulling numbers into your own tools. The tracker is cookieless, so you are in the same consent position as Plausible and Umami.

The server side is a TypeScript application backed by MongoDB, with Redis alongside in the compose setup. That is the first real cost. Plausible needs Postgres and ClickHouse, Umami needs Postgres or MySQL, and most self-hosters already run Postgres for something. Mongo is an extra database to back up and upgrade, and the 512 MB minimum is realistic only for a low-traffic site; give it 1 GB if you have several properties.

The AI assistant needs an API key, which changes the privacy story

Leave the key out and the assistant simply is not there, and Litlyx becomes a straightforward cookieless analytics tool with a modern dashboard. Add the key and every question you ask, plus the aggregated figures needed to answer it, leaves your server. It is aggregated data rather than raw IPs, and OpenAI's API terms say they do not train on it, but "analytics I self-hosted so the numbers never leave my box" is no longer strictly true. For a personal blog that is a non-issue. For a client site with a data processing agreement, you need to write it down.

There is no built-in path to point the assistant at a local model. If that changes, it would remove my main objection.

Events and conversions are where it beats Plausible on paper

Plausible and Umami are page-view tools with events bolted on. Litlyx was designed around events: you define conversions, attach values to them, and the dashboard shows funnels and per-source conversion rates without an add-on. If you run a SaaS landing page and want "trial signups by campaign" without exporting to a spreadsheet, that is more convenient than Plausible goals.

Whether that is enough depends on what you already have. If you send events to a product analytics tool anyway, Litlyx is duplicating it. If Plausible goals have been "fine", they will keep being fine.

Young project, moving fast: what that means for you

Litlyx first shipped in 2024. Its 1,742 stars arrived quickly, which is a good sign for interest and a neutral one for stability. Practically, that means release notes are worth reading before every pull, the schema may change under you, and you should take a mongodump before upgrades. The AGPL-3.0 licence is the same one Plausible uses, so no surprises there, but note the project also sells a hosted plan, and features have a way of landing in the cloud tier first.

A tool you rely on for a year of historical traffic should be one you expect to exist in three years. Litlyx probably will. Umami and Plausible definitely will, and Matomo has 15 years of proof.

What I'd do

If your reason for leaving Google Analytics is privacy and simplicity, install Umami: Postgres, one container, done, and the Plausible vs Umami comparison settles the choice between those two. Read ditching Google Analytics first if you have not made the jump yet.

Choose Litlyx if you specifically want conversion funnels in a self-hosted tool and you are comfortable running Mongo. Run it without the AI key for a month to see whether you miss the assistant. Most people I know who tried it did not, and the ones who did were asking questions a saved dashboard filter would have answered.