py-data-engine 0.3.3

GUI orchestrator for python-based dataframe transform pipelines.

Data Engine

Data Engine is a GUI orchestrator for Python-based dataframe transform pipelines.

It provides:

• a workspace-based runtime for authored flows
• a desktop GUI for operators
• a terminal UI for headless/local operation
• an experimental Rust-backed egui surface
• parquet-first inspect/debug tooling for dataframe outputs

Flows are plain Python modules that declare how source files, settings workbooks, schedules, and manual runs should move through Polars, DuckDB, and file outputs.

What It Is

Data Engine is not just a dataframe library and not just a scheduler. It is the operator/runtime layer around Python-authored flow modules.

The package handles:

• workspace discovery and selection
• daemon ownership and control handoff
• manual, poll, and schedule execution modes
• mirrored output routing
• persisted run/log/state history
• dataframe inspection inside the app

Step functions use normal Python libraries directly. In practice that usually means:

• Polars for dataframe transforms
• DuckDB for SQL-oriented work
• pathlib-style file output

Install

Installer scripts

Use the installer that matches your environment:

• macOS: INSTALL/INSTALL MAC.command
• Windows: INSTALL/INSTALL WINDOWS.bat
• Windows VM / CPU-safe Polars path: INSTALL/INSTALL WINDOWS_VM.bat

Manual install

Base install:

python -m pip install py-data-engine

Editable local install:

python -m pip install -e .

Notebook-authored flow modules (.ipynb) work with the base install. Install the optional notebook extra only when you want Jupyter authoring tools:

python -m pip install -e ".[notebook]"

For contributors:

python -m pip install -e ".[dev]"

Core requirements:

• Python >=3.14
• PySide6 for the desktop GUI
• Textual for the terminal UI

Start The App

Desktop GUI:

data-engine start gui

Experimental Rust egui surface:

data-engine start egui

Terminal UI:

data-engine start tui

You can also launch from module form if needed:

python -m data_engine.ui.cli.app start gui

Headless CLI

data-engine list
data-engine show example_summary
data-engine run --once example_summary
data-engine run

data-engine run starts the automated engine headlessly for discovered automated flows and keeps running until stopped. Use --once to force a single pass instead.

Public API

from data_engine import Flow, FlowContext, discover_flows, load_flow, run

Workspace Model

Data Engine discovers workspaces from a collection root resolved from:

• DATA_ENGINE_WORKSPACE_COLLECTION_ROOT, when explicitly set
• DATA_ENGINE_WORKSPACE_ROOT, when binding directly to one authored workspace
• otherwise the machine-local app settings store

Each immediate child folder containing flow_modules/ is treated as a workspace, for example:

• workspaces/example_workspace/flow_modules/
• workspaces/claims2/flow_modules/

Shared workspace state lives inside each authored workspace:

• workspaces/<workspace_id>/.workspace_state/

Machine-local runtime state lives under the app artifacts root:

• artifacts/workspace_cache/<workspace_id>/
• artifacts/runtime_state/<workspace_id>/

The app's selected workspace and collection-root preference are machine-local settings, not repo-local config.

Flow Shape

from data_engine import Flow
import polars as pl


def read_claims(context):
    return pl.read_excel(context.source.path)


def keep_open(context):
    return context.current.filter(pl.col("status") == "OPEN")


def write_parquet(context):
    output = context.mirror.with_suffix(".parquet")
    context.current.write_parquet(output)
    return output


def build():
    return (
        Flow(group="Claims")
        .watch(
            mode="poll",
            source="../../../example_data/Input/claims_flat",
            interval="5s",
            extensions=[".xlsx", ".xls", ".xlsm"],
            settle=1,
        )
        .mirror(root="../../../example_data/Output/example_mirror")
        .step(read_claims, save_as="raw_df")
        .step(keep_open, use="raw_df", save_as="filtered_df")
        .step(write_parquet, use="filtered_df")
    )

Each flow module exports:

• optional DESCRIPTION
• build() -> Flow

The module filename is the flow identity. Authored flow modules should set Flow(group=...) and let the loader inject the final flow name from the module filename.

Runtime Modes

Flows can run as:

• manual
• poll
• schedule

At a high level:

• manual runs on operator request
• poll watches source inputs for new or changed files
• schedule runs on a time-based cadence

Batch Helpers

For batch-oriented flows, use Flow.collect(...) plus either Flow.map(...) or Flow.step_each(...).

from data_engine import Flow


def validate_workbook(context, file_ref):
    return {
        "name": file_ref.name,
        "path": file_ref.path,
        "ok": file_ref.exists(),
    }


def build():
    return (
        Flow(group="Claims")
        .watch(mode="schedule", run_as="batch", interval="15m", source="../../../example_data/Input/claims_flat")
        .collect([".xlsx"])
        .map(validate_workbook)
    )

Flow.collect(...) returns a Batch of FileRef items. Flow.map(...) applies one callable to each item and returns a new Batch. Flow.step_each(...) is the equivalent readability-first alias.

Flow API

• Flow(group=...)
• .watch(mode="manual", source=None, run_as="individual")
• .watch(mode="poll", source=..., interval=..., extensions=None, settle=1, run_as="individual")
• .watch(mode="schedule", interval=..., source=None, run_as="individual" | "batch")
• .watch(mode="schedule", time="HH:MM", source=None, run_as="individual" | "batch")
• .watch(mode="schedule", time=["08:15", "14:45"], source=..., run_as="individual" | "batch")
• .mirror(root=...)
• .step(fn, use=None, save_as=None, label=None)
• .collect(extensions, root=None, recursive=False, use=None, save_as=None, label=None)
• .map(fn, use=None, save_as=None, label=None)
• .step_each(fn, use=None, save_as=None, label=None)
• .preview(use=None)
• .run_once()
• .run()
• .show()

step() callables always receive one FlowContext parameter and return the next value for context.current.

map() and step_each() callables accept either (item) or (context, item) and return a mapped Batch.

FlowContext

FlowContext exposes the active run state:

• context.source
• context.mirror
• context.current
• context.objects
• context.metadata
• context.source_metadata()
• context.debug

Useful source helpers:

• context.source.path
• context.source.with_extension(".json")
• context.source.with_suffix(".json")
• context.source.file("notes.json")
• context.source.namespaced_file("notes.json")
• context.source.root_file("lookup.csv")

Useful mirror helpers:

• context.mirror.with_extension(".parquet")
• context.mirror.with_suffix(".parquet")
• context.mirror.file("open_claims.parquet")
• context.mirror.namespaced_file("open_claims.parquet")

use="name" loads context.objects["name"] into context.current before the step runs. save_as="name" stores the returned value into context.objects["name"].

Dataframe Debugging

The app includes a dataframe-first debug pane for saved parquet artifacts.

From a flow step, save a debug dataframe with:

context.debug.save_frame(context.current, name="raw_claims")

That writes:

• a parquet artifact for the dataframe
• companion metadata used by the UI

The desktop GUI can then:

• list saved dataframe artifacts by flow/step/timestamp
• inspect parquet outputs in-app
• preview top N, bottom N, or sampled rows
• filter columns with Excel-style distinct-value popups
• copy one or more selected cells from the table

The inspect modal reuses the same dataframe rendering path.

Notebook Preview

For notebook or REPL-style authoring outside a compiled flow module, preview() is usually the most useful helper:

build().preview(use="raw_df").head(10)
build().preview(use="filtered_df")

preview(use="name") runs the flow until that save_as="name" object exists, then returns the real object without running later steps. preview() is not available from inside compiled flow modules, so use it from an external notebook, REPL, or script while iterating on the flow.

Discovery And Compilation

Starter flow modules live in:

• workspaces/<workspace_id>/flow_modules/
• artifacts/workspace_cache/<workspace_id>/compiled_flow_modules/

Authored flow modules compile into:

• artifacts/workspace_cache/<workspace_id>/compiled_flow_modules/*.py

The runtime loads discovered flows from those compiled modules.

Workspace Layout

• src/data_engine/ Runtime package, operator surfaces, and services
• workspaces/<workspace_id>/flow_modules/ Authored flow sources (.py or .ipynb)
• workspaces/<workspace_id>/.workspace_state/ Shared lease markers and checkpoint parquet snapshots
• artifacts/workspace_cache/<workspace_id>/compiled_flow_modules/ Generated importable flow modules
• artifacts/runtime_state/<workspace_id>/ Machine-local runtime and daemon state
• src/data_engine/docs/ Packaged documentation content

Smoke Data

Generate local smoke data with:

python scripts/generate_smoke_data.py --root . --workspace-id example_workspace --workspace-id claims2

Generated local data and workspaces are intentionally ignored:

• data/
• data2/
• workspaces/

Packaging

Distribution name:

• py-data-engine

Version source of truth:

• src/data_engine/platform/identity.py

Build checks:

python -m build
python -m twine check dist/*

Status

This project is pre-alpha. Internal architecture is still moving quickly, and backwards compatibility is not a current goal.