Pipelines as
Readable Rules
Replace pages of PySpark and SQL with a few declarative rules that read, join and write across every source in place. Validate them offline, review them in git, trace every result back to its sources, and run them on compute you already have.
| concept | plan |
|---|---|
| depends_on | modified |
| at_risk | re-run |
| risk_report | reused |
Proven in production
Less Glue.
More Logic.
Most pipeline effort goes into code nobody wants to own: extracts, staging copies, self-join loops and the same definition written five times.
Say What.
Not How.
Connect your inputs, declare what each output means, say where it goes. The engine reads each source where it lives, plans the joins and compiles the logic into parallel, distributed jobs. It runs recursive queries faster than optimized PySpark.
Connect, Define, Write
A connection reads the source, rules define the transformation, and an output writes the result. That is the pipeline.
Read and Write in Place
Files, relational databases and warehouses such as Snowflake, Databricks, BigQuery and Redshift, with append or overwrite.
Recursion Without Loops
Supplier tiers, org charts and bills of materials as recursive rules. The engine iterates to a fixed point.
Any Language
Write a rule in SQL, Python, Cypher or Vadalog, or describe it in plain language and let the assistant draft it for you to review.
Ship new pipelines without new glue code. Retire the self-join loops. Keep logic readable after people move on.
Git In.
Results Out.
The open-source px CLI turns your rules, connections and schema into plain files. Check them offline with px validate, review changes in a pull request, preview what will change with px plan, and ship with px apply from your CI.
Git-Native CLI
Diff a change, review it in a PR, roll back with git. Apache-2.0 licensed.
Plan Before Apply
px validate checks schema and structure offline. px plan previews every create, update and delete, plus which results will re-run.
Schedules and Webhooks
Run rules on a schedule or when data changes, write results back, or trigger them from your orchestrator over REST or Python.
Runs on Your Compute
On Databricks, Kubernetes, or in local mode for development. No new database to stand up.
Treat pipelines like the rest of your code. Keep the orchestrator and clusters you have.
From Repo to
Running Pipeline
Scaffold
Run px init to create a workspace of plain files in your repo.
Connect and Write Rules
Connect sources, write the logic as rules, and say where results go.
Plan and Apply
Validate offline, review the plan in a PR, then apply from your laptop or CI.
Run and Schedule
Run on demand, on a schedule, or from your orchestrator, on compute you already run.
Jobs You'd Rather
Not Maintain
Multi-Tier Exposure
Kora Steel is disrupted. One pipeline finds Lyon, Turin and Gdańsk at risk through every supplier tier and writes the list to your lakehouse.
See the data engine →Cross-System Joins
Join a CRM in Postgres with payments in Snowflake and a graph in Neo4j, without a staging copy.
See cross-database processing →Pipeline Consolidation
Rewrite long, overlapping transformation jobs as shared rules, so the next pipeline reuses definitions you already have instead of copying code again.
See faster ETL →Process Without Copies
Keep sensitive data in the systems that govern it and read it where it lives, with no copy.
See in-place processing →“With Prometheux, we achieved a speed-up of over 250% on critical tasks involving complex reasoning over knowledge graphs with hundreds of millions of nodes and billions of edges.”
Financial institution
“Prometheux technology powers highly scalable and explainable reasoning over some of the world's largest knowledge graphs.”
Accelerating Neurosymbolic AI with RAPIDS and Vadalog Parallel
Questions,
Answered
Do I have to learn a new language?
No. Write rules in plain language and let the AI assistant draft them for you to review, or write them in SQL, Cypher or Python. Vadalog, a readable declarative logic language, is there when you want full control, and every option shares the same concepts and lineage.
Does it replace my orchestrator?
No. Prometheux defines and runs the transformation logic. Rules can run on a schedule or when data changes, or be triggered from your orchestrator through the REST API or Python SDK.
Can AI write the pipelines for us?
It can draft them. The assistant writes Vadalog from a plain-English description, translates existing SQL, and describes any rule in plain English. What you get is a readable rule you review, version in git and run deterministically, not generated code to reverse-engineer.
Where does it run?
On your Databricks clusters, Kubernetes, or local mode for development. Rules compile to parallel, distributed jobs that read each source in place.
How do I version and deploy rules?
With the open-source px CLI. Your workspace is plain files in git: validate offline, preview changes with px plan, apply them with px apply, and run the same steps in CI.
What lineage do we get?
Follow every result back through the concepts and sources it depends on, not just table to table. Turn provenance on for a run and every output fact records the rule that derived it and the source facts it used, including across recursive steps.
Prometheux For…
Shrink Your
Next Pipeline
Bring a PySpark or SQL job you'd rather not maintain. We'll rewrite it as rules with you and run it on your own cluster.
