The billing unit decides your self-hosted customer data platform

On 13 October 2026 Helium Ventures announced it has acquired Jitsu, an open-source customer data platform and Segment alternative. Jitsu collects event data from websites, apps and servers and sends it to data warehouses and business tools in real time. It was founded in 2020, went through Y Combinator's Summer 2020 batch and raised $2 million in seed funding led by Costanoa Ventures.

That is the news, and the press release tells it better than I can. If you run your own stack, the part that matters is the billing unit Jitsu escapes. Segment, owned by Twilio since 2020, bills on monthly tracked users. Jitsu bills per active event, and its code is MIT-licensed and free to self-host.

What a self-hosted customer data platform actually does

A customer data platform is a pipe with three jobs. It collects events from your site, your app and your server: a page view, a button click, an order marked as paid. It transforms them in flight, which means renaming fields, dropping junk and removing personal data you should not store. Then it delivers them to your warehouse and to the tools that need them.

Self-hosted means the pipe runs on your hardware and the events land on disks you control. A few years ago that was an experiment. Today it is a normal option.

The repository at github.com/jitsucom/jitsu is MIT-licensed, has 5,106 stars, is not archived and was last pushed on 13 October 2026, according to the GitHub API on that day. The services that take events in and load them into the warehouse are written in Go. The compose file in the repo starts ClickHouse, Postgres, MongoDB and Redpanda as a Kafka-compatible queue, and Helium's release describes a Kubernetes operator for production.

The billing unit: per tracked person or per active event

Every event pipeline charges for something. The two common units are the person and the event.

With per-person billing, a tracked user is anyone, logged in or anonymous, who fires one event in the month. A visitor who arrives from a search, reads one page and leaves counts the same as a customer who pays you. Your bill grows with traffic, whether or not revenue grows with it.

Twilio's own Segment Connections pricing page, which says "Pricing current as of August 2026", lists these numbers:

• Free: $0, up to 1,000 visitors a month.
• Team: starting at $120 a month, including 10,000 monthly tracked users.
• Between 10,000 and 25,000 tracked users: +$12 per additional 1,000.
• Between 25,000 and 100,000: +$11 per additional 1,000.
• Above 100,000: +$10 per additional 1,000.
• Business: custom volume.

Here is my own arithmetic on those bands for a site with 50,000 monthly tracked users: $120 + 15 × $12 + 25 × $11 = $575 a month. Twilio does not publish that figure. It is my sum on its published bands, so check it against your own quote.

With per-event billing, the bill counts events and ignores how many people sent them. Helium says the Jitsu Cloud free tier covers 200,000 active events a month, and that the Business plan is $99 a month including 2 million active events. Helium also says existing Jitsu Cloud plans and pricing are unchanged.

The two units do not convert into each other without your own data. Spread across 50,000 people, 200,000 events is 4 events each and 2 million is 40 events each. A blog where most visitors load one or two pages looks very different from an app that fires an event on every tap.

So do not trust a calculator, including mine. Send the same events to both systems for one full month, then read your own two bills.

Self-hosting changes the unit again. Helium says Jitsu remains free to self-host with no usage limits, so you pay for a server, disk space and your own hours. Count the hours honestly.

What to check before you self-host an event pipeline

An event you lose is gone for good, so check four things before you send real traffic. One note first: I have not migrated my own tracking to Jitsu or benchmarked it. What follows comes from the repository, its README and Helium's release.

Durability and replay

Ask what happens when your warehouse is down for an hour, and how you run old events again after you fix a broken transform. Helium says Jitsu 2.14, released in July, adds event reprocessing. Test it yourself: stop the destination, send a known number of events, start it again and count the rows.

Who can reach the data

An event stream holds email addresses, IP addresses and order values. Helium lists audit logs, expiring API tokens and a maintenance mode in the same release. Look at the defaults too. The repo's docker/README.md documents a seeded login of [email protected] / admin123 and service tokens that default to dev-auth-key. Change all of them before any port faces the internet.

Where the data lands

The reason to run the pipe yourself is that events go to your warehouse and your disk. That makes backups your job. The compose file keeps Postgres data in ./data/postgres beside the compose file, so know which disk that is, how full it is and whether your backup run includes it.

The operational surface

Count the moving parts. The compose stack runs the Jitsu services console, rotor, bulker and ingest, plus Postgres, MongoDB, ClickHouse and Redpanda. A fifth service, syncctl, runs connector syncs, and the README says it needs Kubernetes and exits without it.

Helium says 2.14 gives self-hosted teams a Kubernetes operator, dedicated function servers and Kubernetes-native sync scheduling. On a single Docker host, plan for event streaming first.

What the acquisition changes and what it does not

Helium says Jitsu remains MIT-licensed and free to self-host, and that the releases since the acquisition expand production self-hosting. It lists new destinations including Resend, SendGrid, Statsig, DuckDB/MotherDuck and Xero, plus a Management API and CLI.

Read that for what it is. The promise to preserve the open-source foundation is the acquirer's own statement in its own press release. It is intent, not a contractual guarantee. The release also says the pace of releases "has increased since the acquisition", which tells you the deal closed before it was announced.

What you can rely on today is the MIT licence on the code as published. A later owner can change the licence for later versions. Nobody can take back a version that is already out under MIT. So the version you already run is the safe part, and the roadmap is the part to watch.

In my post on Deno and Cloudflare I argued that an acquisition promise is not an exit plan. Here that means three small habits. Set RELEASE_TAG to a specific version instead of latest, keep a copy of the source for that tag, and read the licence file before each upgrade. As with open-weight models, owning software only counts if you can run it yourself.

How to start this week

Clone the repository and read docker/README.md first. The command it gives is:

docker compose -f ./docker/docker-compose.yml up --force-recreate

According to the README, the console answers on http://localhost:3000 and the ingestion endpoint on port 3049. It also labels this setup as for exploration and development only.

Before you start it, set your own SEED_USER_PASSWORD, database passwords and service tokens. The README says to do that with environment variables or a .env.local file, and to leave .env alone.

If you do not want to run a server yet, Helium says the Jitsu Cloud free tier covers 200,000 active events a month.

Either way, run it alongside your current pipeline so nothing breaks. Leave your existing tracking code in place, add the second one beside it and send the same events to both for a full billing month. The two tools are not identical, so check each destination and transform you depend on before you switch anything off.

If you are wiring AI agents into your stack, Helium says Jitsu also ships an MCP server. Treat an agent like any other client. Give it a token that expires, and read the audit log after it has run.

If you want a hand

None of this needs a specialist, but it does need time. If you would rather have someone set up an event pipeline and harden it for you, that is work I take on. The details are on the services page.

Fiverr: hiteshsaini459 · Upwork: hiteshsaini25