Export to Portable Format

Export to Portable Format packages a project’s workflows as a self-contained Python program that runs on DuckDB with no connection to PlaidCloud. It produces a portable copy of the project’s logic and data that runs offline, on its own.

It is a one-way export. The generated package is a point-in-time snapshot; it does not sync back, and re-exporting after changes produces a fresh package.

Best-effort fidelity

The export reproduces SQL-based transform steps exactly. Proprietary and side-effecting operations — allocations, dimension loads, notifications, document and report steps, scripts — are exported as stubs: their original configuration is preserved and, where the output table was materialized, its data is bundled as a snapshot, but the step itself does not re-run. Each step’s fidelity is recorded in COVERAGE.md inside the package. For guaranteed, fully-recomputed results, run the project in PlaidCloud.

Exporting a Project

1. Open Analyze and select the Projects tab.
2. Select a single project.
3. Open the Actions menu (or right-click the project) and choose Export to Portable Format — it sits beside Export Project.
4. In the dialog:
   • Document Account — the document account the package is written to.
   • Folder Path — the folder within that account, for example /exports.
   • Include data — when checked (the default), source and snapshot tables are bundled as Parquet so the package runs without any external data. Uncheck it to ship only the code and schema contracts, which you then populate yourself.
5. Click Export. When it finishes, a message confirms where the package was written and how many tables were snapshotted.

The action requires access to the project and to read its tables — the same level of access needed to export table data today.

What’s in the Package

The export is a .zip containing:

• pipeline_<workflow>.py — one script per workflow, with the steps in execution order.
• config.py — engine selection (DuckDB by default) and the project schema name.
• _runtime.py — a small engine shim; the only dependency beyond DuckDB.
• stubs.py — no-op functions for non-portable operations, each carrying the original step configuration.
• data/ — Parquet snapshots of source and frozen tables (when Include data was checked).
• requirements.txt — the Python dependencies (just duckdb).
• README.md, COVERAGE.md, RUN_ORDER.md — how to run it, per-step fidelity, and the order to run the workflow scripts in.

Running the Package

1. Unzip the package and create a Python environment:

   python -m venv .venv && source .venv/bin/activate
   pip install -r requirements.txt

2. Run each workflow script, in the order listed in RUN_ORDER.md:

   python pipeline_<workflow>.py

Each script builds its tables in a local DuckDB database (export.duckdb by default). SQL transform steps are recomputed live; stubbed steps print a notice and leave their snapshotted output in place. To point the package at your own warehouse instead of DuckDB, set ENGINE in config.py.

Related

• Archive a project — a full, restorable backup that stays within PlaidCloud
• Manage tables and views — the data layer the export snapshots