Run

Records

Run history, schedules, agent sessions and code revisions are small rows; node outputs are files (above). The rows go wherever EXUVIA_DB points:

EXUVIA_DB
unsetSQLite at .exuvia/exuvia.db in each project. Nothing to set up; a hub export carries it.
sqlite:///<path>one SQLite file for every project the process serves
postgresql://…Postgres (pip install 'exuvia[postgres]')

In a shared database every row carries its project's scope, EXUVIA_DB_SCOPE + the project id, so several hubs (one per user, say) can share one database by each setting their own EXUVIA_DB_SCOPE. A hub export then leaves the records out, since they already live in the database, and deleting a project deletes its rows. A project's identity there is its directory name, so moving a standalone project to a differently named folder starts a fresh history. One process at a time serves a given project.

The SQL is one dialect SQLite and Postgres both accept (? placeholders, on conflict upserts, window functions). Another backend, such as an HTTP SQL service like Cloudflare D1, is a class implementing exuvia.db.Database: query, transaction, columns, close. A project database written by an older exuvia is upgraded in place when it opens.