Every analyst. Every engineer. One canvas.

The strongest visual ETL engine. Build pipelines in your browser, move them through staging to production, and run them on your own servers. One flat price.

Start your 30-day trial

orders · nightly refresh Draft Unsaved v7 Staging v7 · Production v7 Succeeded

Postgres

Input

482,910 rows

CSV Files

Input

12,480 rows

Filter

Preparation

311,204 rows

Data Cleanse

Preparation

12,480 rows

Join

309,877 rows

S3 File

Output

309,877 rows

Data Out Messages Metadata Join · no run neededJoin · 50-row sampleProduction v7

order_idInt64

placed_atTimestamp

customerUtf8

amountDecimal128

regionUtf8

is_priorityBoolean

order_id	customer	region	amount
1043	Redwood Labs	West	1,284.00
1044	Bellhaven Co	North	312.50
1045	Orchard Group	West	9,940.10
1046	Kestrel Freight	South	2,410.00

02:00Started on Production v7
02:01Postgres read 482,910 rows
02:02Filter kept 311,204 of 482,910 rows
02:03Join matched 309,877 rows
02:04S3 File wrote 309,877 rows. Succeeded.

Who it's for

Analysts build it. Engineers ship it. IT signs it off.

Everyone works on the same pipeline in the same browser, so the person who asked for a change can open it, test it and see exactly what it does.

Analysts & business teams

Build it yourself, without a ticket or a licence to wait for.

A 50-row preview of every step, one click of Test away
Formulas in plain SQL, with [bracket] column names if you prefer them
Excel in, and formatted Excel reports and per-person emails out

The Monday report, off the desktop

Data engineers

Production pipelines without the glue code.

S3, Parquet, Iceberg and seven warehouses and query engines
Schedules, file-arrival triggers and Flows that branch on failure
Staging and production versions, with per-node telemetry on every run

Data-lake pipelines, without the glue

IT, security & platform Single sign-on, directory sign-in, eight roles and an audit log of every change. Data scientists Your own Python as a step in the pipeline, in a sandboxed container with pandas and polars ready. Everyone else Results arrive as an Excel report, a Google Sheet or a Salesforce record, and an account of their own costs nothing.

Built for production

The discipline engineers expect, on a canvas analysts can read.

Everything a pipeline needs once people depend on it: versions, schedules, orchestration, alerts and a record of every run.

One pipeline is a graph of rows. A Flow is a graph of pipelines.

Run pipelines in sequence or side by side, branch on the outcome, and alert someone only when something fails.

Rolling back is an edit, not an incident.

Every publish is a numbered version nobody can change, and every run is stamped with the one it ran.

v7StagingProduction
v6Restore
v5Restore

Four ways a run starts.

Each one lands in the same queue, the same history and the same telemetry.

On a schedule
When a file lands
Once per variant
As a step in a Flow

Every run answers what happened.

Rows and time for every node, and peak memory and queue wait for every run.

Postgres482,910
Join309,877
S3 File309,877

One expression language, and it is SQL.

Filters, formulas and validation rules all share it, and [bracket] column names still work.

CASE WHEN [score] >= 90 THEN 'A'
     WHEN [score] >= 80 THEN 'B'
     ELSE 'C' END

The engine

100% Rust. No JVM. No garbage collector.

500,000,000rows, one server

80 node pipeline 8 vCPU, 32 GB of memory 33m 55s end to end

One measured run: half a billion rows read from a Parquet file on S3, sent through 75 steps of filters, formulas and column changes, then totalled and checked. Not one row was lost or duplicated, and more than 120 GB spilled to disk rather than running out of memory.

Many established visual ETL tools run on the Java virtual machine, which is why keeping one alive under load means heap flags, per-component memory settings and a lot of time reading stack traces. That is the standing cost of a garbage-collected data engine.

Trinium's engine is Rust from end to end, with no virtual machine to size and no collector to pause a run in the middle of the night. Every node runs at once, rows stream between them in batches, and the work that cannot stream spills to disk inside one memory budget instead of taking the process down.

That is why one ordinary server does work that usually needs a cluster, and why the price can be flat.

Read how execution works

Pricing

Priced per server. Never per person.

Pro counts one thing, the servers you run, and Enterprise counts nothing at all. Everyone in the company can have an account on either, and a bigger machine never costs more.

Put in your own numbers Compare the editions

Noseat licencesEveryone in the company can have an account.

Noper-core licencesA 64-core box is still one server.

Nometered runsRun a pipeline every minute if you want to.

Nometered rowsVolume is a capacity question, not a billing one.

Nodes & connectors

Every tool the job needs, already on the canvas.

On the way in: files, databases, warehouses, lakehouse tables and SaaS APIs. In the middle: cleansing, joins, parsing, reshaping, fuzzy matching, validation and reconciliation. On the way out: the same systems again, plus formatted Excel reports and email.

There are no add-on packs and no marketplace, and everything listed ships in the product.

Browse every node

Connects to

PostgreSQL

MySQL

SQL Server

MongoDB

Snowflake

BigQuery

Amazon Redshift

Databricks

ClickHouse

Trino / Starburst

Amazon Athena

Apache Iceberg

Amazon S3

Azure Blob Storage

Google Cloud Storage

SFTP

REST API

Google Sheets

Salesforce

HubSpot

Jira

Zendesk

SMTP (Email)

Self-hosted, properly

Your data never reaches us. Neither does your telemetry.

Read the security overview

Runs in your network. Trinium is a set of containers on your servers, beside a Postgres database you own, and nothing leaves except what your pipelines send.
Licences verify offline. Your key is checked against a public key built into the binary, so there is no licence server and no usage reporting.
Credentials stay encrypted. Secrets are encrypted at rest, and a pipeline holds only the name of a connection, never its password.
Nobody gets locked out. If a licence lapses, new runs stop and everything else keeps working. Nothing is deleted.

Bring your hardest pipeline.

Rebuild one real pipeline of yours during the trial. It takes an afternoon, and it settles the question better than anything on this page.

Start your 30-day trial Browse every node