SQLite Sprig 0.1.0a1 — one record, with its context

SQLite Sprig takes a local SQLite file, a root table and a SQL condition, then creates a selected branch with the parents and ordinary schema it needs.

python sqlite_sprig.py app.sqlite3 --root cards --where "id=103" --out card-103.sqlite3

The source is opened read-only. Selected rows keep their writable values; ordinary tables, indexes, views, triggers and defaults stay in the output. Triggers are restored after data loading. Required parents do not bring in their unrelated child branches. JSON reports the result, and an existing destination is refused.

Choose a download:

• sqlite_sprig.py: standalone CLI, using Python 3.11+ with SQLite 3.37+ and no third-party packages.
• sqlite_sprig-0.1.0a1-py3-none-any.whl: install locally with python -m pip install ./sqlite_sprig-0.1.0a1-py3-none-any.whl; then use sqlite-sprig or python -m sqlite_sprig.
• sqlite-sprig-demo.zip: unzip and open sqlite-sprig-demo/index.html to explore three authored examples and their actual databases. The same files can be served locally with python -m http.server --bind 127.0.0.1 from that folder. Browser testing covered the HTTP path; direct file-origin automation was unavailable.
• The source distribution, editable source in the repository, demo image and SHA256SUMS are also available.

The pictured card retains 11 of 24 authored rows. This is a row-selection example, not a size or speed claim.

Validation: 16 local behavior/CLI tests, a clean wheel installation outside the checkout, actual view/default/trigger and no-overwrite probes, and Linux/Windows/macOS CI on Python 3.11 and 3.13. Desktop and narrow browser layouts, keyboard selection, command copying and a real SQLite download were checked. Hosted macOS is not physical-device validation.

Alpha boundaries: this copies actual data without masking. It uses declared foreign keys and does not infer application relationships or minimize a failing case. Virtual tables and external SQLite extensions are outside this preview. Aggregate views and application invariants still need case-specific checks. See the README for the complete contract.

A small internal two-agent comparison had both the direct-script and prototype arms succeed in roughly the same time. Net savings and external adoption have not been demonstrated. A concrete schema or selection case is more useful feedback than a general benchmark claim.