A visual canvas
Read, filter, join, aggregate, pivot, write: each block carries its parameters, and the graph is validated before it runs.
← Overview
Low-code pipelines
Draw the pipeline. The code is yours.
You assemble the steps on a canvas; graal writes the matching Pandas or PySpark code. That code is readable, versioned and runs outside graal: on a laptop, in your CI, on another cluster.
A valid pipeline, exported to PySpark from the editor. The same menu offers Pandas; the button next to it turns the pipeline into a job.
Real graal console interface, not retouched; fictional demonstration data (tenant energie-demo — people, projects and tokens are invented).
Key capabilities
From drawing to code, nothing lost
A preview at every step
Click a block and the pipeline runs up to it on a sample, showing the rows it produces.
Runnable export
Pandas for modest volumes, PySpark to scale out. The code can be reviewed, tested and run without graal.
Built-in quality checks
Not null, uniqueness, value ranges, regular expressions, freshness: a failing check stops the run or warns, according to your rule.
Pipelines described in plain language
You describe the processing in a sentence; graal proposes the graph, which you review and adjust before saving it.
Versions and Git
Every save creates a version of the graph, committed to the project’s Git repository. You compare, you roll back.
How it works
From canvas to scheduled job
Step 01
Draw
You place the blocks, connect them and check the data preview at every step.
Step 02
Check
You add quality checks wherever the data has to keep its promises. Their report comes with every run.
Step 03
Industrialize
The graph becomes a schedulable job that fits into a workflow. Or you export the code and deploy it elsewhere.
The graph is the source, the code is its translation
A graal pipeline is a directed graph: each block declares its inputs, its parameters and its output schema. The engine validates the graph, then translates it into a complete Pandas or PySpark script, dependencies included. There is no hidden execution: what runs in graal is what you export.
A pipeline reads like code
Because the generated code is readable, it fits into your existing practices: code review, unit tests, continuous integration. Your data engineers keep control over what goes to production, and your analysts build without waiting.
The European Data Act prohibits switching charges between data processing service providers from 12 January 2027 (Article 29). Exported code and open formats make that switch concrete.
Standards and integrations
Formats you keep
- Pandas
- PySpark
- Python
- Parquet
- Apache Iceberg
- SQL
- Git
Governance
Reversible by design
- Exported code is yours and runs without a graal license
- No credential in the generated code, only references to secrets
- Separate permissions to edit, export and run a pipeline, project by project
Frequently asked questions
What is left if we leave graal?
The Pandas or PySpark code of your pipelines, their Git history, and your data in Parquet and Iceberg. Nothing needs converting.
Do we need to code?
Not to draw. The generated code is meant to be reviewed by your data engineers: it is the reference.
Pandas or PySpark?
The same graph exports to both. Pandas suits data that fits in memory; PySpark distributes the work across your cluster.
Watch a pipeline become a job
From drawing to code export, then to a scheduled job: it is one of the acts of the demonstration.