You want funnels, retention, and user profiles for your product. What you don't want is to operate a column database to get them. Until now, Tell's answer was "run ClickHouse" — great at billions of events, overkill on a single VPS.

Now the whole analytics suite also runs on files. ClickHouse didn't go anywhere — you have a second first-class store, and the choice is yours.

One binary, your files

Point Tell at a directory and start sending events. Data lands as Apache Arrow files — an open columnar format any data tool can read — and every query runs directly against them. No database server, no connection strings, nothing else to install.

Every report works — funnels, retention, lifecycle, user paths, people profiles, audiences, revenue, group analytics. Same features, same numbers
Verified, not promised — a differential test suite runs each metric against both the file store and ClickHouse and requires identical results
Your data stays yours — plain files on your disk, in a format DuckDB, pandas, or a Python script can open tomorrow

Or point it at a bucket

Set mode = "s3" and the file store becomes a shared one — AWS S3, MinIO, or Cloudflare R2.

Stateless collectors — ingest nodes upload straight to the bucket and keep nothing local. Kill one, start another, no data to migrate
Readers scan the same bucket — the dashboard queries what collectors wrote, seconds later
Hot local, durable remote — local+upload keeps recent data on disk for fast queries while every file is also uploaded for safekeeping
You can see it working — tell status shows upload backlog, and the server checks the bucket is reachable before it starts taking traffic

It cleans up after itself

File stores usually mean a cron job and a shell script. Not here.

Background compaction — small files merge into bigger ones automatically, so queries stay fast as data grows
Retention windows — tell it to keep 90 days and older partitions are deleted for you
On your schedule — tell maintenance runs a one-shot compaction and cleanup pass without restarting the server

ClickHouse or files — your call

Two stores, equal citizens. Every feature works on both, and the differential suite holds them to identical answers — so the choice is about your deployment, not your feature list.

Choose ClickHouse for proven scale — billions of events, a mature operational ecosystem, and blazing queries over deep history
Choose files for zero moving parts — one binary, an open on-disk format, and object-storage durability without running a database
Switch whenever — it's a config line, and your boards, funnels, and audiences don't change

Either way, the same binary scales from a laptop to a fleet of stateless collectors sharing one bucket. And don't assume files are the slow option — early numbers hold up better than we expected. We're benchmarking, not calling a winner.